Add JAX typography, xector primitives, deferred effect chains, and GPU streaming

- Add JAX text rendering with font atlas, styled text placement, and typography primitives
- Add xector (element-wise/reduction) operations library and sexp effects
- Add deferred effect chain fusion for JIT-compiled effect pipelines
- Expand drawing primitives with font management, alignment, shadow, and outline
- Add interpreter support for function-style define and require
- Add GPU persistence mode and hardware decode support to streaming
- Add new sexp effects: cell_pattern, halftone, mosaic, and derived definitions
- Add path registry for asset resolution
- Add integration, primitives, and xector tests

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

path_registry.py

@@ -0,0 +1,477 @@
+"""
+Path Registry - Maps human-friendly paths to content-addressed IDs.
+
+This module provides a bidirectional mapping between:
+- Human-friendly paths (e.g., "effects/ascii_fx_zone.sexp")
+- Content-addressed IDs (IPFS CIDs or SHA3-256 hashes)
+
+The registry is useful for:
+- Looking up effects by their friendly path name
+- Resolving cids back to the original path for debugging
+- Maintaining a stable naming scheme across cache updates
+
+Storage:
+- Uses the existing item_types table in the database (path column)
+- Caches in Redis for fast lookups across distributed workers
+
+The registry uses a system actor (@system@local) for global path mappings,
+allowing effects to be resolved by path without requiring user context.
+"""
+
+import logging
+import os
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Dict, List, Optional, Tuple
+from dataclasses import dataclass
+
+logger = logging.getLogger(__name__)
+
+# System actor for global path mappings (effects, recipes, analyzers)
+SYSTEM_ACTOR = "@system@local"
+
+
+@dataclass
+class PathEntry:
+    """A registered path with its content-addressed ID."""
+    path: str  # Human-friendly path (relative or normalized)
+    cid: str   # Content-addressed ID (IPFS CID or hash)
+    content_type: str  # Type: "effect", "recipe", "analyzer", etc.
+    actor_id: str = SYSTEM_ACTOR  # Owner (system for global)
+    description: Optional[str] = None
+    created_at: float = 0.0
+
+
+class PathRegistry:
+    """
+    Registry for mapping paths to content-addressed IDs.
+
+    Uses the existing item_types table for persistence and Redis
+    for fast lookups in distributed Celery workers.
+    """
+
+    def __init__(self, redis_client=None):
+        self._redis = redis_client
+        self._redis_path_to_cid_key = "artdag:path_to_cid"
+        self._redis_cid_to_path_key = "artdag:cid_to_path"
+
+    def _run_async(self, coro):
+        """Run async coroutine from sync context."""
+        import asyncio
+
+        try:
+            loop = asyncio.get_running_loop()
+            import threading
+            result = [None]
+            error = [None]
+
+            def run_in_thread():
+                try:
+                    new_loop = asyncio.new_event_loop()
+                    asyncio.set_event_loop(new_loop)
+                    try:
+                        result[0] = new_loop.run_until_complete(coro)
+                    finally:
+                        new_loop.close()
+                except Exception as e:
+                    error[0] = e
+
+            thread = threading.Thread(target=run_in_thread)
+            thread.start()
+            thread.join(timeout=30)
+            if error[0]:
+                raise error[0]
+            return result[0]
+        except RuntimeError:
+            try:
+                loop = asyncio.get_event_loop()
+            except RuntimeError:
+                loop = asyncio.new_event_loop()
+                asyncio.set_event_loop(loop)
+            return loop.run_until_complete(coro)
+
+    def _normalize_path(self, path: str) -> str:
+        """Normalize a path for consistent storage."""
+        # Remove leading ./ or /
+        path = path.lstrip('./')
+        # Normalize separators
+        path = path.replace('\\', '/')
+        # Remove duplicate slashes
+        while '//' in path:
+            path = path.replace('//', '/')
+        return path
+
+    def register(
+        self,
+        path: str,
+        cid: str,
+        content_type: str = "effect",
+        actor_id: str = SYSTEM_ACTOR,
+        description: Optional[str] = None,
+    ) -> PathEntry:
+        """
+        Register a path -> cid mapping.
+
+        Args:
+            path: Human-friendly path (e.g., "effects/ascii_fx_zone.sexp")
+            cid: Content-addressed ID (IPFS CID or hash)
+            content_type: Type of content ("effect", "recipe", "analyzer")
+            actor_id: Owner (default: system for global mappings)
+            description: Optional description
+
+        Returns:
+            The created PathEntry
+        """
+        norm_path = self._normalize_path(path)
+        now = datetime.now(timezone.utc).timestamp()
+
+        entry = PathEntry(
+            path=norm_path,
+            cid=cid,
+            content_type=content_type,
+            actor_id=actor_id,
+            description=description,
+            created_at=now,
+        )
+
+        # Store in database (item_types table)
+        self._save_to_db(entry)
+
+        # Update Redis cache
+        self._update_redis_cache(norm_path, cid)
+
+        logger.info(f"Registered path '{norm_path}' -> {cid[:16]}...")
+        return entry
+
+    def _save_to_db(self, entry: PathEntry):
+        """Save entry to database using item_types table."""
+        import database
+
+        async def save():
+            import asyncpg
+            conn = await asyncpg.connect(database.DATABASE_URL)
+            try:
+                # Ensure cache_item exists
+                await conn.execute(
+                    "INSERT INTO cache_items (cid) VALUES ($1) ON CONFLICT DO NOTHING",
+                    entry.cid
+                )
+                # Insert or update item_type with path
+                await conn.execute(
+                    """
+                    INSERT INTO item_types (cid, actor_id, type, path, description)
+                    VALUES ($1, $2, $3, $4, $5)
+                    ON CONFLICT (cid, actor_id, type, path) DO UPDATE SET
+                        description = COALESCE(EXCLUDED.description, item_types.description)
+                    """,
+                    entry.cid, entry.actor_id, entry.content_type, entry.path, entry.description
+                )
+            finally:
+                await conn.close()
+
+        try:
+            self._run_async(save())
+        except Exception as e:
+            logger.warning(f"Failed to save path registry to DB: {e}")
+
+    def _update_redis_cache(self, path: str, cid: str):
+        """Update Redis cache with mapping."""
+        if self._redis:
+            try:
+                self._redis.hset(self._redis_path_to_cid_key, path, cid)
+                self._redis.hset(self._redis_cid_to_path_key, cid, path)
+            except Exception as e:
+                logger.warning(f"Failed to update Redis cache: {e}")
+
+    def get_cid(self, path: str, content_type: str = None) -> Optional[str]:
+        """
+        Get the cid for a path.
+
+        Args:
+            path: Human-friendly path
+            content_type: Optional type filter
+
+        Returns:
+            The cid, or None if not found
+        """
+        norm_path = self._normalize_path(path)
+
+        # Try Redis first (fast path)
+        if self._redis:
+            try:
+                val = self._redis.hget(self._redis_path_to_cid_key, norm_path)
+                if val:
+                    return val.decode() if isinstance(val, bytes) else val
+            except Exception as e:
+                logger.warning(f"Redis lookup failed: {e}")
+
+        # Fall back to database
+        return self._get_cid_from_db(norm_path, content_type)
+
+    def _get_cid_from_db(self, path: str, content_type: str = None) -> Optional[str]:
+        """Get cid from database using item_types table."""
+        import database
+
+        async def get():
+            import asyncpg
+            conn = await asyncpg.connect(database.DATABASE_URL)
+            try:
+                if content_type:
+                    row = await conn.fetchrow(
+                        "SELECT cid FROM item_types WHERE path = $1 AND type = $2",
+                        path, content_type
+                    )
+                else:
+                    row = await conn.fetchrow(
+                        "SELECT cid FROM item_types WHERE path = $1",
+                        path
+                    )
+                return row["cid"] if row else None
+            finally:
+                await conn.close()
+
+        try:
+            result = self._run_async(get())
+            # Update Redis cache if found
+            if result and self._redis:
+                self._update_redis_cache(path, result)
+            return result
+        except Exception as e:
+            logger.warning(f"Failed to get from DB: {e}")
+            return None
+
+    def get_path(self, cid: str) -> Optional[str]:
+        """
+        Get the path for a cid.
+
+        Args:
+            cid: Content-addressed ID
+
+        Returns:
+            The path, or None if not found
+        """
+        # Try Redis first
+        if self._redis:
+            try:
+                val = self._redis.hget(self._redis_cid_to_path_key, cid)
+                if val:
+                    return val.decode() if isinstance(val, bytes) else val
+            except Exception as e:
+                logger.warning(f"Redis lookup failed: {e}")
+
+        # Fall back to database
+        return self._get_path_from_db(cid)
+
+    def _get_path_from_db(self, cid: str) -> Optional[str]:
+        """Get path from database using item_types table."""
+        import database
+
+        async def get():
+            import asyncpg
+            conn = await asyncpg.connect(database.DATABASE_URL)
+            try:
+                row = await conn.fetchrow(
+                    "SELECT path FROM item_types WHERE cid = $1 AND path IS NOT NULL ORDER BY created_at LIMIT 1",
+                    cid
+                )
+                return row["path"] if row else None
+            finally:
+                await conn.close()
+
+        try:
+            result = self._run_async(get())
+            # Update Redis cache if found
+            if result and self._redis:
+                self._update_redis_cache(result, cid)
+            return result
+        except Exception as e:
+            logger.warning(f"Failed to get from DB: {e}")
+            return None
+
+    def list_by_type(self, content_type: str, actor_id: str = None) -> List[PathEntry]:
+        """
+        List all entries of a given type.
+
+        Args:
+            content_type: Type to filter by ("effect", "recipe", etc.)
+            actor_id: Optional actor filter (None = all, SYSTEM_ACTOR = global)
+
+        Returns:
+            List of PathEntry objects
+        """
+        import database
+
+        async def list_entries():
+            import asyncpg
+            conn = await asyncpg.connect(database.DATABASE_URL)
+            try:
+                if actor_id:
+                    rows = await conn.fetch(
+                        """
+                        SELECT cid, path, type, actor_id, description,
+                               EXTRACT(EPOCH FROM created_at) as created_at
+                        FROM item_types
+                        WHERE type = $1 AND actor_id = $2 AND path IS NOT NULL
+                        ORDER BY path
+                        """,
+                        content_type, actor_id
+                    )
+                else:
+                    rows = await conn.fetch(
+                        """
+                        SELECT cid, path, type, actor_id, description,
+                               EXTRACT(EPOCH FROM created_at) as created_at
+                        FROM item_types
+                        WHERE type = $1 AND path IS NOT NULL
+                        ORDER BY path
+                        """,
+                        content_type
+                    )
+                return [
+                    PathEntry(
+                        path=row["path"],
+                        cid=row["cid"],
+                        content_type=row["type"],
+                        actor_id=row["actor_id"],
+                        description=row["description"],
+                        created_at=row["created_at"] or 0,
+                    )
+                    for row in rows
+                ]
+            finally:
+                await conn.close()
+
+        try:
+            return self._run_async(list_entries())
+        except Exception as e:
+            logger.warning(f"Failed to list from DB: {e}")
+            return []
+
+    def delete(self, path: str, content_type: str = None) -> bool:
+        """
+        Delete a path registration.
+
+        Args:
+            path: The path to delete
+            content_type: Optional type filter
+
+        Returns:
+            True if deleted, False if not found
+        """
+        norm_path = self._normalize_path(path)
+
+        # Get cid for Redis cleanup
+        cid = self.get_cid(norm_path, content_type)
+
+        # Delete from database
+        deleted = self._delete_from_db(norm_path, content_type)
+
+        # Clean up Redis
+        if deleted and cid and self._redis:
+            try:
+                self._redis.hdel(self._redis_path_to_cid_key, norm_path)
+                self._redis.hdel(self._redis_cid_to_path_key, cid)
+            except Exception as e:
+                logger.warning(f"Failed to clean up Redis: {e}")
+
+        return deleted
+
+    def _delete_from_db(self, path: str, content_type: str = None) -> bool:
+        """Delete from database."""
+        import database
+
+        async def delete():
+            import asyncpg
+            conn = await asyncpg.connect(database.DATABASE_URL)
+            try:
+                if content_type:
+                    result = await conn.execute(
+                        "DELETE FROM item_types WHERE path = $1 AND type = $2",
+                        path, content_type
+                    )
+                else:
+                    result = await conn.execute(
+                        "DELETE FROM item_types WHERE path = $1",
+                        path
+                    )
+                return "DELETE" in result
+            finally:
+                await conn.close()
+
+        try:
+            return self._run_async(delete())
+        except Exception as e:
+            logger.warning(f"Failed to delete from DB: {e}")
+            return False
+
+    def register_effect(
+        self,
+        path: str,
+        cid: str,
+        description: Optional[str] = None,
+    ) -> PathEntry:
+        """
+        Convenience method to register an effect.
+
+        Args:
+            path: Effect path (e.g., "effects/ascii_fx_zone.sexp")
+            cid: IPFS CID of the effect file
+            description: Optional description
+
+        Returns:
+            The created PathEntry
+        """
+        return self.register(
+            path=path,
+            cid=cid,
+            content_type="effect",
+            actor_id=SYSTEM_ACTOR,
+            description=description,
+        )
+
+    def get_effect_cid(self, path: str) -> Optional[str]:
+        """
+        Get CID for an effect by path.
+
+        Args:
+            path: Effect path
+
+        Returns:
+            IPFS CID or None
+        """
+        return self.get_cid(path, content_type="effect")
+
+    def list_effects(self) -> List[PathEntry]:
+        """List all registered effects."""
+        return self.list_by_type("effect", actor_id=SYSTEM_ACTOR)
+
+
+# Singleton instance
+_registry: Optional[PathRegistry] = None
+
+
+def get_path_registry() -> PathRegistry:
+    """Get the singleton path registry instance."""
+    global _registry
+    if _registry is None:
+        import redis
+        from urllib.parse import urlparse
+
+        redis_url = os.environ.get('REDIS_URL', 'redis://localhost:6379/5')
+        parsed = urlparse(redis_url)
+        redis_client = redis.Redis(
+            host=parsed.hostname or 'localhost',
+            port=parsed.port or 6379,
+            db=int(parsed.path.lstrip('/') or 0),
+            socket_timeout=5,
+            socket_connect_timeout=5
+        )
+
+        _registry = PathRegistry(redis_client=redis_client)
+    return _registry
+
+
+def reset_path_registry():
+    """Reset the singleton (for testing)."""
+    global _registry
+    _registry = None

sexp_effects/derived.sexp

@@ -0,0 +1,206 @@
+;; Derived Operations
+;;
+;; These are built from true primitives using S-expressions.
+;; Load with: (require "derived")
+
+;; =============================================================================
+;; Math Helpers (derivable from where + basic ops)
+;; =============================================================================
+
+;; Absolute value
+(define (abs x) (where (< x 0) (- x) x))
+
+;; Minimum of two values
+(define (min2 a b) (where (< a b) a b))
+
+;; Maximum of two values
+(define (max2 a b) (where (> a b) a b))
+
+;; Clamp x to range [lo, hi]
+(define (clamp x lo hi) (max2 lo (min2 hi x)))
+
+;; Square of x
+(define (sq x) (* x x))
+
+;; Linear interpolation: a*(1-t) + b*t
+(define (lerp a b t) (+ (* a (- 1 t)) (* b t)))
+
+;; Smooth interpolation between edges
+(define (smoothstep edge0 edge1 x)
+  (let ((t (clamp (/ (- x edge0) (- edge1 edge0)) 0 1)))
+    (* t (* t (- 3 (* 2 t))))))
+
+;; =============================================================================
+;; Channel Shortcuts (derivable from channel primitive)
+;; =============================================================================
+
+;; Extract red channel as xector
+(define (red frame) (channel frame 0))
+
+;; Extract green channel as xector
+(define (green frame) (channel frame 1))
+
+;; Extract blue channel as xector
+(define (blue frame) (channel frame 2))
+
+;; Convert to grayscale xector (ITU-R BT.601)
+(define (gray frame)
+  (+ (* (red frame) 0.299)
+     (* (green frame) 0.587)
+     (* (blue frame) 0.114)))
+
+;; Alias for gray
+(define (luminance frame) (gray frame))
+
+;; =============================================================================
+;; Coordinate Generators (derivable from iota + repeat/tile)
+;; =============================================================================
+
+;; X coordinate for each pixel [0, width)
+(define (x-coords frame) (tile (iota (width frame)) (height frame)))
+
+;; Y coordinate for each pixel [0, height)
+(define (y-coords frame) (repeat (iota (height frame)) (width frame)))
+
+;; Normalized X coordinate [0, 1]
+(define (x-norm frame) (/ (x-coords frame) (max2 1 (- (width frame) 1))))
+
+;; Normalized Y coordinate [0, 1]
+(define (y-norm frame) (/ (y-coords frame) (max2 1 (- (height frame) 1))))
+
+;; Distance from frame center for each pixel
+(define (dist-from-center frame)
+  (let* ((cx (/ (width frame) 2))
+         (cy (/ (height frame) 2))
+         (dx (- (x-coords frame) cx))
+         (dy (- (y-coords frame) cy)))
+    (sqrt (+ (sq dx) (sq dy)))))
+
+;; Normalized distance from center [0, ~1]
+(define (dist-norm frame)
+  (let ((d (dist-from-center frame)))
+    (/ d (max2 1 (βmax d)))))
+
+;; =============================================================================
+;; Cell/Grid Operations (derivable from floor + basic math)
+;; =============================================================================
+
+;; Cell row index for each pixel
+(define (cell-row frame cell-size) (floor (/ (y-coords frame) cell-size)))
+
+;; Cell column index for each pixel
+(define (cell-col frame cell-size) (floor (/ (x-coords frame) cell-size)))
+
+;; Number of cell rows
+(define (num-rows frame cell-size) (floor (/ (height frame) cell-size)))
+
+;; Number of cell columns
+(define (num-cols frame cell-size) (floor (/ (width frame) cell-size)))
+
+;; Flat cell index for each pixel
+(define (cell-indices frame cell-size)
+  (+ (* (cell-row frame cell-size) (num-cols frame cell-size))
+     (cell-col frame cell-size)))
+
+;; Total number of cells
+(define (num-cells frame cell-size)
+  (* (num-rows frame cell-size) (num-cols frame cell-size)))
+
+;; X position within cell [0, cell-size)
+(define (local-x frame cell-size) (mod (x-coords frame) cell-size))
+
+;; Y position within cell [0, cell-size)
+(define (local-y frame cell-size) (mod (y-coords frame) cell-size))
+
+;; Normalized X within cell [0, 1]
+(define (local-x-norm frame cell-size)
+  (/ (local-x frame cell-size) (max2 1 (- cell-size 1))))
+
+;; Normalized Y within cell [0, 1]
+(define (local-y-norm frame cell-size)
+  (/ (local-y frame cell-size) (max2 1 (- cell-size 1))))
+
+;; =============================================================================
+;; Fill Operations (derivable from iota)
+;; =============================================================================
+
+;; Xector of n zeros
+(define (zeros n) (* (iota n) 0))
+
+;; Xector of n ones
+(define (ones n) (+ (zeros n) 1))
+
+;; Xector of n copies of val
+(define (fill val n) (+ (zeros n) val))
+
+;; Xector of zeros matching x's length
+(define (zeros-like x) (* x 0))
+
+;; Xector of ones matching x's length
+(define (ones-like x) (+ (zeros-like x) 1))
+
+;; =============================================================================
+;; Pooling (derivable from group-reduce)
+;; =============================================================================
+
+;; Pool a channel by cell index
+(define (pool-channel chan cell-idx num-cells)
+  (group-reduce chan cell-idx num-cells "mean"))
+
+;; Pool red channel to cells
+(define (pool-red frame cell-size)
+  (pool-channel (red frame)
+                (cell-indices frame cell-size)
+                (num-cells frame cell-size)))
+
+;; Pool green channel to cells
+(define (pool-green frame cell-size)
+  (pool-channel (green frame)
+                (cell-indices frame cell-size)
+                (num-cells frame cell-size)))
+
+;; Pool blue channel to cells
+(define (pool-blue frame cell-size)
+  (pool-channel (blue frame)
+                (cell-indices frame cell-size)
+                (num-cells frame cell-size)))
+
+;; Pool grayscale to cells
+(define (pool-gray frame cell-size)
+  (pool-channel (gray frame)
+                (cell-indices frame cell-size)
+                (num-cells frame cell-size)))
+
+;; =============================================================================
+;; Blending (derivable from math)
+;; =============================================================================
+
+;; Additive blend
+(define (blend-add a b) (clamp (+ a b) 0 255))
+
+;; Multiply blend (normalized)
+(define (blend-multiply a b) (* (/ a 255) b))
+
+;; Screen blend
+(define (blend-screen a b) (- 255 (* (/ (- 255 a) 255) (- 255 b))))
+
+;; Overlay blend
+(define (blend-overlay a b)
+  (where (< a 128)
+         (* 2 (/ (* a b) 255))
+         (- 255 (* 2 (/ (* (- 255 a) (- 255 b)) 255)))))
+
+;; =============================================================================
+;; Simple Effects (derivable from primitives)
+;; =============================================================================
+
+;; Invert a channel (255 - c)
+(define (invert-channel c) (- 255 c))
+
+;; Binary threshold
+(define (threshold-channel c thresh) (where (> c thresh) 255 0))
+
+;; Reduce to n levels
+(define (posterize-channel c levels)
+  (let ((step (/ 255 (- levels 1))))
+    (* (round (/ c step)) step)))

sexp_effects/effects/ascii_art.sexp

@@ -5,7 +5,7 @@
   :params (
     (char_size :type int :default 8 :range [4 32])
     (alphabet :type string :default "standard")
-    (color_mode :type string :default "color" :desc ""color", "mono", "invert", or any color name/hex")
+    (color_mode :type string :default "color" :desc "color, mono, invert, or any color name/hex")
     (background_color :type string :default "black" :desc "background color name/hex")
     (invert_colors :type int :default 0 :desc "swap foreground and background colors")
     (contrast :type float :default 1.5 :range [1 3])

sexp_effects/effects/cell_pattern.sexp

@@ -0,0 +1,65 @@
+;; Cell Pattern effect - custom patterns within cells
+;;
+;; Demonstrates building arbitrary per-cell visuals from primitives.
+;; Uses local coordinates within cells to draw patterns scaled by luminance.
+
+(require-primitives "xector")
+
+(define-effect cell_pattern
+  :params (
+    (cell-size :type int :default 16 :range [8 48] :desc "Cell size")
+    (pattern :type string :default "diagonal" :desc "Pattern: diagonal, cross, ring")
+  )
+  (let* (
+         ;; Pool to get cell colors
+         (pooled (pool-frame frame cell-size))
+         (cell-r (nth pooled 0))
+         (cell-g (nth pooled 1))
+         (cell-b (nth pooled 2))
+         (cell-lum (α/ (nth pooled 3) 255))
+
+         ;; Cell indices for each pixel
+         (cell-idx (cell-indices frame cell-size))
+
+         ;; Look up cell values for each pixel
+         (pix-r (gather cell-r cell-idx))
+         (pix-g (gather cell-g cell-idx))
+         (pix-b (gather cell-b cell-idx))
+         (pix-lum (gather cell-lum cell-idx))
+
+         ;; Local position within cell [0, 1]
+         (lx (local-x-norm frame cell-size))
+         (ly (local-y-norm frame cell-size))
+
+         ;; Pattern mask based on pattern type
+         (mask
+           (cond
+             ;; Diagonal lines - thickness based on luminance
+             ((= pattern "diagonal")
+              (let* ((diag (αmod (α+ lx ly) 0.25))
+                     (thickness (α* pix-lum 0.125)))
+                (α< diag thickness)))
+
+             ;; Cross pattern
+             ((= pattern "cross")
+              (let* ((cx (αabs (α- lx 0.5)))
+                     (cy (αabs (α- ly 0.5)))
+                     (thickness (α* pix-lum 0.25)))
+                (αor (α< cx thickness) (α< cy thickness))))
+
+             ;; Ring pattern
+             ((= pattern "ring")
+              (let* ((dx (α- lx 0.5))
+                     (dy (α- ly 0.5))
+                     (dist (αsqrt (α+ (α² dx) (α² dy))))
+                     (target (α* pix-lum 0.4))
+                     (thickness 0.05))
+                (α< (αabs (α- dist target)) thickness)))
+
+             ;; Default: solid
+             (else (α> pix-lum 0)))))
+
+    ;; Apply mask: show cell color where mask is true, black elsewhere
+    (rgb (where mask pix-r 0)
+         (where mask pix-g 0)
+         (where mask pix-b 0))))

sexp_effects/effects/echo.sexp

@@ -6,10 +6,10 @@
     (num_echoes :type int :default 4 :range [1 20])
     (decay :type float :default 0.5 :range [0 1])
   )
-  (let* ((buffer (state-get 'buffer (list)))
+  (let* ((buffer (state-get "buffer" (list)))
          (new-buffer (take (cons frame buffer) (+ num_echoes 1))))
     (begin
-      (state-set 'buffer new-buffer)
+      (state-set "buffer" new-buffer)
       ;; Blend frames with decay
       (if (< (length new-buffer) 2)
           frame

sexp_effects/effects/halftone.sexp

@@ -0,0 +1,49 @@
+;; Halftone/dot effect - built from primitive xector operations
+;;
+;; Uses:
+;;   pool-frame     - downsample to cell luminances
+;;   cell-indices   - which cell each pixel belongs to
+;;   gather         - look up cell value for each pixel
+;;   local-x/y-norm - position within cell [0,1]
+;;   where          - conditional per-pixel
+
+(require-primitives "xector")
+
+(define-effect halftone
+  :params (
+    (cell-size :type int :default 12 :range [4 32] :desc "Size of halftone cells")
+    (dot-scale :type float :default 0.9 :range [0.1 1.0] :desc "Max dot radius")
+    (invert :type bool :default false :desc "Invert (white dots on black)")
+  )
+  (let* (
+         ;; Pool frame to get luminance per cell
+         (pooled (pool-frame frame cell-size))
+         (cell-lum (nth pooled 3))  ; luminance is 4th element
+
+         ;; For each output pixel, get its cell index
+         (cell-idx (cell-indices frame cell-size))
+
+         ;; Get cell luminance for each pixel
+         (pixel-lum (α/ (gather cell-lum cell-idx) 255))
+
+         ;; Position within cell, normalized to [-0.5, 0.5]
+         (lx (α- (local-x-norm frame cell-size) 0.5))
+         (ly (α- (local-y-norm frame cell-size) 0.5))
+
+         ;; Distance from cell center (0 at center, ~0.7 at corners)
+         (dist (αsqrt (α+ (α² lx) (α² ly))))
+
+         ;; Radius based on luminance (brighter = bigger dot)
+         (radius (α* (if invert (α- 1 pixel-lum) pixel-lum)
+                     (α* dot-scale 0.5)))
+
+         ;; Is this pixel inside the dot?
+         (inside (α< dist radius))
+
+         ;; Output color
+         (fg (if invert 255 0))
+         (bg (if invert 0 255))
+         (out (where inside fg bg)))
+
+    ;; Grayscale output
+    (rgb out out out)))

sexp_effects/effects/mosaic.sexp

@@ -0,0 +1,30 @@
+;; Mosaic effect - built from primitive xector operations
+;;
+;; Uses:
+;;   pool-frame  - downsample to cell averages
+;;   cell-indices - which cell each pixel belongs to
+;;   gather       - look up cell value for each pixel
+
+(require-primitives "xector")
+
+(define-effect mosaic
+  :params (
+    (cell-size :type int :default 16 :range [4 64] :desc "Size of mosaic cells")
+  )
+  (let* (
+         ;; Pool frame to get average color per cell (returns r,g,b,lum xectors)
+         (pooled (pool-frame frame cell-size))
+         (cell-r (nth pooled 0))
+         (cell-g (nth pooled 1))
+         (cell-b (nth pooled 2))
+
+         ;; For each output pixel, get its cell index
+         (cell-idx (cell-indices frame cell-size))
+
+         ;; Gather: look up cell color for each pixel
+         (out-r (gather cell-r cell-idx))
+         (out-g (gather cell-g cell-idx))
+         (out-b (gather cell-b cell-idx)))
+
+    ;; Reconstruct frame
+    (rgb out-r out-g out-b)))

sexp_effects/effects/outline.sexp

@@ -5,9 +5,9 @@
   :params (
     (thickness :type int :default 2 :range [1 10])
     (threshold :type int :default 100 :range [20 300])
-    (color :type list :default (list 0 0 0)
+    (color :type list :default (list 0 0 0))
+    (fill_mode :type string :default "original")
   )
-  (fill_mode "original"))
   (let* ((edge-img (image:edge-detect frame (/ threshold 2) threshold))
          (dilated (if (> thickness 1)
                       (dilate edge-img thickness)

sexp_effects/effects/strobe.sexp

@@ -5,12 +5,12 @@
   :params (
     (frame_rate :type int :default 12 :range [1 60])
   )
-  (let* ((held (state-get 'held nil))
-         (held-until (state-get 'held-until 0))
+  (let* ((held (state-get "held" nil))
+         (held-until (state-get "held-until" 0))
          (frame-duration (/ 1 frame_rate)))
     (if (or (core:is-nil held) (>= t held-until))
         (begin
-          (state-set 'held (copy frame))
-          (state-set 'held-until (+ t frame-duration))
+          (state-set "held" (copy frame))
+          (state-set "held-until" (+ t frame-duration))
           frame)
         held)))

sexp_effects/effects/trails.sexp

@@ -5,16 +5,16 @@
   :params (
     (persistence :type float :default 0.8 :range [0 0.99])
   )
-  (let* ((buffer (state-get 'buffer nil))
+  (let* ((buffer (state-get "buffer" nil))
          (current frame))
     (if (= buffer nil)
         (begin
-          (state-set 'buffer (copy frame))
+          (state-set "buffer" (copy frame))
           frame)
         (let* ((faded (blending:blend-images buffer
                                     (make-image (image:width frame) (image:height frame) (list 0 0 0))
                                     (- 1 persistence)))
                (result (blending:blend-mode faded current "lighten")))
           (begin
-            (state-set 'buffer result)
+            (state-set "buffer" result)
             result)))))

sexp_effects/effects/xector_feathered_blend.sexp

@@ -0,0 +1,44 @@
+;; Feathered blend - blend two same-size frames with distance-based falloff
+;; Center shows overlay, edges show background, with smooth transition
+
+(require-primitives "xector")
+
+(define-effect xector_feathered_blend
+  :params (
+    (inner-radius :type float :default 0.3 :range [0 1] :desc "Radius where overlay is 100% (fraction of size)")
+    (fade-width :type float :default 0.2 :range [0 0.5] :desc "Width of fade region (fraction of size)")
+    (overlay :type frame :default nil :desc "Frame to blend in center")
+  )
+  (let* (
+         ;; Get normalized distance from center (0 at center, ~1 at corners)
+         (dist (dist-from-center frame))
+         (max-dist (βmax dist))
+         (dist-norm (α/ dist max-dist))
+
+         ;; Calculate blend factor:
+         ;; - 1.0 when dist-norm < inner-radius (fully overlay)
+         ;; - 0.0 when dist-norm > inner-radius + fade-width (fully background)
+         ;; - linear ramp between
+         (t (α/ (α- dist-norm inner-radius) fade-width))
+         (blend (α- 1 (αclamp t 0 1)))
+         (inv-blend (α- 1 blend))
+
+         ;; Background channels
+         (bg-r (red frame))
+         (bg-g (green frame))
+         (bg-b (blue frame)))
+
+    (if (nil? overlay)
+        ;; No overlay - visualize the blend mask
+        (let ((vis (α* blend 255)))
+          (rgb vis vis vis))
+
+        ;; Blend overlay with background using the mask
+        (let* ((ov-r (red overlay))
+               (ov-g (green overlay))
+               (ov-b (blue overlay))
+               ;; lerp: bg * (1-blend) + overlay * blend
+               (r-out (α+ (α* bg-r inv-blend) (α* ov-r blend)))
+               (g-out (α+ (α* bg-g inv-blend) (α* ov-g blend)))
+               (b-out (α+ (α* bg-b inv-blend) (α* ov-b blend))))
+          (rgb r-out g-out b-out)))))

sexp_effects/effects/xector_grain.sexp

@@ -0,0 +1,34 @@
+;; Film grain effect using xector operations
+;; Demonstrates random xectors and mixing scalar/xector math
+
+(require-primitives "xector")
+
+(define-effect xector_grain
+  :params (
+    (intensity :type float :default 0.2 :range [0 1] :desc "Grain intensity")
+    (colored :type bool :default false :desc "Use colored grain")
+  )
+  (let* (
+         ;; Extract channels
+         (r (red frame))
+         (g (green frame))
+         (b (blue frame))
+
+         ;; Generate noise xector(s)
+         ;; randn-x generates normal distribution noise
+         (grain-amount (* intensity 50)))
+
+    (if colored
+        ;; Colored grain: different noise per channel
+        (let* ((nr (randn-x frame 0 grain-amount))
+               (ng (randn-x frame 0 grain-amount))
+               (nb (randn-x frame 0 grain-amount)))
+          (rgb (αclamp (α+ r nr) 0 255)
+               (αclamp (α+ g ng) 0 255)
+               (αclamp (α+ b nb) 0 255)))
+
+        ;; Monochrome grain: same noise for all channels
+        (let ((n (randn-x frame 0 grain-amount)))
+          (rgb (αclamp (α+ r n) 0 255)
+               (αclamp (α+ g n) 0 255)
+               (αclamp (α+ b n) 0 255))))))

sexp_effects/effects/xector_inset_blend.sexp

@@ -0,0 +1,57 @@
+;; Inset blend - fade a smaller frame into a larger background
+;; Uses distance-based alpha for smooth transition (no hard edges)
+
+(require-primitives "xector")
+
+(define-effect xector_inset_blend
+  :params (
+    (x :type int :default 0 :desc "X position of inset")
+    (y :type int :default 0 :desc "Y position of inset")
+    (fade-width :type int :default 50 :desc "Width of fade region in pixels")
+    (overlay :type frame :default nil :desc "The smaller frame to inset")
+  )
+  (let* (
+         ;; Get dimensions
+         (bg-h (first (list (nth (list (red frame)) 0))))  ;; TODO: need image:height
+         (bg-w bg-h)  ;; placeholder
+
+         ;; For now, create a simple centered circular blend
+         ;; Distance from center of overlay position
+         (cx (+ x (/ (- bg-w (* 2 x)) 2)))
+         (cy (+ y (/ (- bg-h (* 2 y)) 2)))
+
+         ;; Get coordinates as xectors
+         (px (x-coords frame))
+         (py (y-coords frame))
+
+         ;; Distance from center
+         (dx (α- px cx))
+         (dy (α- py cy))
+         (dist (αsqrt (α+ (α* dx dx) (α* dy dy))))
+
+         ;; Inner radius (fully overlay) and outer radius (fully background)
+         (inner-r (- (/ bg-w 2) x fade-width))
+         (outer-r (- (/ bg-w 2) x))
+
+         ;; Blend factor: 1.0 inside inner-r, 0.0 outside outer-r, linear between
+         (t (α/ (α- dist inner-r) fade-width))
+         (blend (α- 1 (αclamp t 0 1)))
+
+         ;; Extract channels from both frames
+         (bg-r (red frame))
+         (bg-g (green frame))
+         (bg-b (blue frame)))
+
+    ;; If overlay provided, blend it
+    (if overlay
+        (let* ((ov-r (red overlay))
+               (ov-g (green overlay))
+               (ov-b (blue overlay))
+               ;; Linear blend: result = bg * (1-blend) + overlay * blend
+               (r-out (α+ (α* bg-r (α- 1 blend)) (α* ov-r blend)))
+               (g-out (α+ (α* bg-g (α- 1 blend)) (α* ov-g blend)))
+               (b-out (α+ (α* bg-b (α- 1 blend)) (α* ov-b blend))))
+          (rgb r-out g-out b-out))
+        ;; No overlay - just show the blend mask for debugging
+        (let ((mask-vis (α* blend 255)))
+          (rgb mask-vis mask-vis mask-vis)))))

sexp_effects/effects/xector_threshold.sexp

@@ -0,0 +1,27 @@
+;; Threshold effect using xector operations
+;; Demonstrates where (conditional select) and β (reduction) for normalization
+
+(require-primitives "xector")
+
+(define-effect xector_threshold
+  :params (
+    (threshold :type float :default 0.5 :range [0 1] :desc "Brightness threshold (0-1)")
+    (invert :type bool :default false :desc "Invert the threshold")
+  )
+  (let* (
+         ;; Get grayscale luminance as xector
+         (luma (gray frame))
+
+         ;; Normalize to 0-1 range
+         (luma-norm (α/ luma 255))
+
+         ;; Create boolean mask: pixels above threshold
+         (mask (if invert
+                   (α< luma-norm threshold)
+                   (α>= luma-norm threshold)))
+
+         ;; Use where to select: white (255) if above threshold, black (0) if below
+         (out (where mask 255 0)))
+
+    ;; Output as grayscale (same value for R, G, B)
+    (rgb out out out)))

sexp_effects/effects/xector_vignette.sexp

@@ -0,0 +1,36 @@
+;; Vignette effect using xector operations
+;; Demonstrates α (element-wise) and β (reduction) patterns
+
+(require-primitives "xector")
+
+(define-effect xector_vignette
+  :params (
+    (strength :type float :default 0.5 :range [0 1])
+    (radius :type float :default 1.0 :range [0.5 2])
+  )
+  (let* (
+         ;; Get normalized distance from center for each pixel
+         (dist (dist-from-center frame))
+
+         ;; Calculate max distance (corner distance)
+         (max-dist (* (βmax dist) radius))
+
+         ;; Calculate brightness factor per pixel: 1 - (dist/max-dist * strength)
+         ;; Using explicit α operators
+         (factor (α- 1 (α* (α/ dist max-dist) strength)))
+
+         ;; Clamp factor to [0, 1]
+         (factor (αclamp factor 0 1))
+
+         ;; Extract channels as xectors
+         (r (red frame))
+         (g (green frame))
+         (b (blue frame))
+
+         ;; Apply factor to each channel (implicit element-wise via Xector operators)
+         (r-out (* r factor))
+         (g-out (* g factor))
+         (b-out (* b factor)))
+
+    ;; Combine back to frame
+    (rgb r-out g-out b-out)))

sexp_effects/interpreter.py

@@ -156,11 +156,21 @@ class Interpreter:
                 if form == 'define':
                     name = expr[1]
                     if _is_symbol(name):
+                        # Simple define: (define name value)
                         value = self.eval(expr[2], env)
                         self.global_env.set(name.name, value)
                         return value
+                    elif isinstance(name, list) and len(name) >= 1 and _is_symbol(name[0]):
+                        # Function define: (define (fn-name args...) body)
+                        # Desugars to: (define fn-name (lambda (args...) body))
+                        fn_name = name[0].name
+                        params = [p.name if _is_symbol(p) else p for p in name[1:]]
+                        body = expr[2]
+                        fn = Lambda(params, body, env)
+                        self.global_env.set(fn_name, fn)
+                        return fn
                     else:
-                        raise SyntaxError(f"define requires symbol, got {name}")
+                        raise SyntaxError(f"define requires symbol or (name args...), got {name}")
 
                 # Define-effect
                 if form == 'define-effect':
@@ -276,6 +286,10 @@ class Interpreter:
                 if form == 'require-primitives':
                     return self._eval_require_primitives(expr, env)
 
+                # require - load .sexp file into current scope
+                if form == 'require':
+                    return self._eval_require(expr, env)
+
             # Function call
             fn = self.eval(head, env)
             args = [self.eval(arg, env) for arg in expr[1:]]
@@ -488,6 +502,61 @@ class Interpreter:
         from .primitive_libs import load_primitive_library
         return load_primitive_library(name, path)
 
+    def _eval_require(self, expr: Any, env: Environment) -> Any:
+        """
+        Evaluate require: load a .sexp file and evaluate its definitions.
+
+        Syntax:
+            (require "derived")           ; loads derived.sexp from sexp_effects/
+            (require "path/to/file.sexp") ; loads from explicit path
+
+        Definitions from the file are added to the current environment.
+        """
+        for lib_expr in expr[1:]:
+            if _is_symbol(lib_expr):
+                lib_name = lib_expr.name
+            else:
+                lib_name = lib_expr
+
+            # Find the .sexp file
+            sexp_path = self._find_sexp_file(lib_name)
+            if sexp_path is None:
+                raise ValueError(f"Cannot find sexp file: {lib_name}")
+
+            # Parse and evaluate the file
+            content = parse_file(sexp_path)
+
+            # Evaluate all top-level expressions
+            if isinstance(content, list) and content and isinstance(content[0], list):
+                for e in content:
+                    self.eval(e, env)
+            else:
+                self.eval(content, env)
+
+        return None
+
+    def _find_sexp_file(self, name: str) -> Optional[str]:
+        """Find a .sexp file by name."""
+        # Try various locations
+        candidates = [
+            # Explicit path
+            name,
+            name + '.sexp',
+            # In sexp_effects directory
+            Path(__file__).parent / f'{name}.sexp',
+            Path(__file__).parent / name,
+            # In effects directory
+            Path(__file__).parent / 'effects' / f'{name}.sexp',
+            Path(__file__).parent / 'effects' / name,
+        ]
+
+        for path in candidates:
+            p = Path(path) if not isinstance(path, Path) else path
+            if p.exists() and p.is_file():
+                return str(p)
+
+        return None
+
     def _eval_ascii_fx_zone(self, expr: Any, env: Environment) -> Any:
         """
         Evaluate ascii-fx-zone special form.
@@ -876,8 +945,8 @@ class Interpreter:
         for pname, pdefault in effect.params.items():
             value = params.get(pname)
             if value is None:
-                # Evaluate default if it's an expression (list)
-                if isinstance(pdefault, list):
+                # Evaluate default if it's an expression (list) or a symbol (like 'nil')
+                if isinstance(pdefault, list) or _is_symbol(pdefault):
                     value = self.eval(pdefault, env)
                 else:
                     value = pdefault

sexp_effects/parser.py

@@ -71,7 +71,8 @@ class Tokenizer:
     STRING = re.compile(r'"(?:[^"\\]|\\.)*"')
     NUMBER = re.compile(r'-?(?:\d+\.?\d*|\.\d+)(?:[eE][+-]?\d+)?')
     KEYWORD = re.compile(r':[a-zA-Z_][a-zA-Z0-9_-]*')
-    SYMBOL = re.compile(r'[a-zA-Z_*+\-><=/!?][a-zA-Z0-9_*+\-><=/!?.:]*')
+    # Symbol pattern includes Greek letters α (alpha) and β (beta) for xector operations
+    SYMBOL = re.compile(r'[a-zA-Z_*+\-><=/!?αβ²λ][a-zA-Z0-9_*+\-><=/!?.:αβ²λ]*')
 
     def __init__(self, text: str):
         self.text = text

sexp_effects/primitive_libs/drawing.py

@@ -1,126 +1,680 @@
 """
 Drawing Primitives Library
 
-Draw shapes, text, and characters on images.
+Draw shapes, text, and characters on images with sophisticated text handling.
+
+Text Features:
+- Font loading from files or system fonts
+- Text measurement and fitting
+- Alignment (left/center/right, top/middle/bottom)
+- Opacity for fade effects
+- Multi-line text support
+- Shadow and outline effects
 """
 import numpy as np
 import cv2
 from PIL import Image, ImageDraw, ImageFont
+import os
+import glob as glob_module
+from typing import Optional, Tuple, List, Union
 
 
-# Default font (will be loaded lazily)
-_default_font = None
+# =============================================================================
+# Font Management
+# =============================================================================
+
+# Font cache: (path, size) -> font object
+_font_cache = {}
+
+# Common system font directories
+FONT_DIRS = [
+    "/usr/share/fonts",
+    "/usr/local/share/fonts",
+    "~/.fonts",
+    "~/.local/share/fonts",
+    "/System/Library/Fonts",  # macOS
+    "/Library/Fonts",  # macOS
+    "C:/Windows/Fonts",  # Windows
+]
+
+# Default fonts to try (in order of preference)
+DEFAULT_FONTS = [
+    "DejaVuSans.ttf",
+    "DejaVuSansMono.ttf",
+    "Arial.ttf",
+    "Helvetica.ttf",
+    "FreeSans.ttf",
+    "LiberationSans-Regular.ttf",
+]
 
 
-def _get_default_font(size=16):
-    """Get default font, creating if needed."""
-    global _default_font
-    if _default_font is None or _default_font.size != size:
+def _find_font_file(name: str) -> Optional[str]:
+    """Find a font file by name in system directories."""
+    # If it's already a full path
+    if os.path.isfile(name):
+        return name
+
+    # Expand user paths
+    expanded = os.path.expanduser(name)
+    if os.path.isfile(expanded):
+        return expanded
+
+    # Search in font directories
+    for font_dir in FONT_DIRS:
+        font_dir = os.path.expanduser(font_dir)
+        if not os.path.isdir(font_dir):
+            continue
+
+        # Direct match
+        direct = os.path.join(font_dir, name)
+        if os.path.isfile(direct):
+            return direct
+
+        # Recursive search
+        for root, dirs, files in os.walk(font_dir):
+            for f in files:
+                if f.lower() == name.lower():
+                    return os.path.join(root, f)
+                # Also match without extension
+                base = os.path.splitext(f)[0]
+                if base.lower() == name.lower():
+                    return os.path.join(root, f)
+
+    return None
+
+
+def _get_default_font(size: int = 24) -> ImageFont.FreeTypeFont:
+    """Get a default font at the given size."""
+    for font_name in DEFAULT_FONTS:
+        path = _find_font_file(font_name)
+        if path:
+            try:
+                return ImageFont.truetype(path, size)
+            except:
+                continue
+
+    # Last resort: PIL default
+    return ImageFont.load_default()
+
+
+def prim_make_font(name_or_path: str, size: int = 24) -> ImageFont.FreeTypeFont:
+    """
+    Load a font by name or path.
+
+    (make-font "Arial" 32)           ; system font by name
+    (make-font "/path/to/font.ttf" 24)  ; font file path
+    (make-font "DejaVuSans" 48)      ; searches common locations
+
+    Returns a font object for use with text primitives.
+    """
+    size = int(size)
+
+    # Check cache
+    cache_key = (name_or_path, size)
+    if cache_key in _font_cache:
+        return _font_cache[cache_key]
+
+    # Find the font file
+    path = _find_font_file(name_or_path)
+    if not path:
+        raise FileNotFoundError(f"Font not found: {name_or_path}")
+
+    # Load and cache
+    font = ImageFont.truetype(path, size)
+    _font_cache[cache_key] = font
+    return font
+
+
+def prim_list_fonts() -> List[str]:
+    """
+    List available system fonts.
+
+    (list-fonts)  ; -> ("Arial.ttf" "DejaVuSans.ttf" ...)
+
+    Returns list of font filenames found in system directories.
+    """
+    fonts = set()
+
+    for font_dir in FONT_DIRS:
+        font_dir = os.path.expanduser(font_dir)
+        if not os.path.isdir(font_dir):
+            continue
+
+        for root, dirs, files in os.walk(font_dir):
+            for f in files:
+                if f.lower().endswith(('.ttf', '.otf', '.ttc')):
+                    fonts.add(f)
+
+    return sorted(fonts)
+
+
+def prim_font_size(font: ImageFont.FreeTypeFont) -> int:
+    """
+    Get the size of a font.
+
+    (font-size my-font)  ; -> 24
+    """
+    return font.size
+
+
+# =============================================================================
+# Text Measurement
+# =============================================================================
+
+def prim_text_size(text: str, font=None, font_size: int = 24) -> Tuple[int, int]:
+    """
+    Measure text dimensions.
+
+    (text-size "Hello" my-font)      ; -> (width height)
+    (text-size "Hello" :font-size 32) ; -> (width height) with default font
+
+    For multi-line text, returns total bounding box.
+    """
+    if font is None:
+        font = _get_default_font(int(font_size))
+    elif isinstance(font, (int, float)):
+        font = _get_default_font(int(font))
+
+    # Create temporary image for measurement
+    img = Image.new('RGB', (1, 1))
+    draw = ImageDraw.Draw(img)
+
+    bbox = draw.textbbox((0, 0), str(text), font=font)
+    width = bbox[2] - bbox[0]
+    height = bbox[3] - bbox[1]
+
+    return (width, height)
+
+
+def prim_text_metrics(font=None, font_size: int = 24) -> dict:
+    """
+    Get font metrics.
+
+    (text-metrics my-font)  ; -> {ascent: 20, descent: 5, height: 25}
+
+    Useful for precise text layout.
+    """
+    if font is None:
+        font = _get_default_font(int(font_size))
+    elif isinstance(font, (int, float)):
+        font = _get_default_font(int(font))
+
+    ascent, descent = font.getmetrics()
+    return {
+        'ascent': ascent,
+        'descent': descent,
+        'height': ascent + descent,
+        'size': font.size,
+    }
+
+
+def prim_fit_text_size(text: str, max_width: int, max_height: int,
+                       font_name: str = None, min_size: int = 8,
+                       max_size: int = 500) -> int:
+    """
+    Calculate font size to fit text within bounds.
+
+    (fit-text-size "Hello World" 400 100)  ; -> 48
+    (fit-text-size "Title" 800 200 :font-name "Arial")
+
+    Returns the largest font size that fits within max_width x max_height.
+    """
+    max_width = int(max_width)
+    max_height = int(max_height)
+    min_size = int(min_size)
+    max_size = int(max_size)
+    text = str(text)
+
+    # Binary search for optimal size
+    best_size = min_size
+    low, high = min_size, max_size
+
+    while low <= high:
+        mid = (low + high) // 2
+
+        if font_name:
+            try:
+                font = prim_make_font(font_name, mid)
+            except:
+                font = _get_default_font(mid)
+        else:
+            font = _get_default_font(mid)
+
+        w, h = prim_text_size(text, font)
+
+        if w <= max_width and h <= max_height:
+            best_size = mid
+            low = mid + 1
+        else:
+            high = mid - 1
+
+    return best_size
+
+
+def prim_fit_font(text: str, max_width: int, max_height: int,
+                  font_name: str = None, min_size: int = 8,
+                  max_size: int = 500) -> ImageFont.FreeTypeFont:
+    """
+    Create a font sized to fit text within bounds.
+
+    (fit-font "Hello World" 400 100)  ; -> font object
+    (fit-font "Title" 800 200 :font-name "Arial")
+
+    Returns a font object at the optimal size.
+    """
+    size = prim_fit_text_size(text, max_width, max_height,
+                               font_name, min_size, max_size)
+
+    if font_name:
         try:
-            _default_font = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSansMono.ttf", size)
+            return prim_make_font(font_name, size)
         except:
-            _default_font = ImageFont.load_default()
-    return _default_font
+            pass
 
+    return _get_default_font(size)
+
+
+# =============================================================================
+# Text Drawing
+# =============================================================================
+
+def prim_text(img: np.ndarray, text: str,
+              x: int = None, y: int = None,
+              width: int = None, height: int = None,
+              font=None, font_size: int = 24, font_name: str = None,
+              color=None, opacity: float = 1.0,
+              align: str = "left", valign: str = "top",
+              fit: bool = False,
+              shadow: bool = False, shadow_color=None, shadow_offset: int = 2,
+              outline: bool = False, outline_color=None, outline_width: int = 1,
+              line_spacing: float = 1.2) -> np.ndarray:
+    """
+    Draw text with alignment, opacity, and effects.
+
+    Basic usage:
+        (text frame "Hello" :x 100 :y 50)
+
+    Centered in frame:
+        (text frame "Title" :align "center" :valign "middle")
+
+    Fit to box:
+        (text frame "Big Text" :x 50 :y 50 :width 400 :height 100 :fit true)
+
+    With fade (for animations):
+        (text frame "Fading" :x 100 :y 100 :opacity 0.5)
+
+    With effects:
+        (text frame "Shadow" :x 100 :y 100 :shadow true)
+        (text frame "Outline" :x 100 :y 100 :outline true :outline-color (0 0 0))
+
+    Args:
+        img: Input frame
+        text: Text to draw
+        x, y: Position (if not specified, uses alignment in full frame)
+        width, height: Bounding box (for fit and alignment within box)
+        font: Font object from make-font
+        font_size: Size if no font specified
+        font_name: Font name to load
+        color: RGB tuple (default white)
+        opacity: 0.0 (invisible) to 1.0 (opaque) for fading
+        align: "left", "center", "right"
+        valign: "top", "middle", "bottom"
+        fit: If true, auto-size font to fit in box
+        shadow: Draw drop shadow
+        shadow_color: Shadow color (default black)
+        shadow_offset: Shadow offset in pixels
+        outline: Draw text outline
+        outline_color: Outline color (default black)
+        outline_width: Outline thickness
+        line_spacing: Multiplier for line height (for multi-line)
+
+    Returns:
+        Frame with text drawn
+    """
+    h, w = img.shape[:2]
+    text = str(text)
+
+    # Default colors
+    if color is None:
+        color = (255, 255, 255)
+    else:
+        color = tuple(int(c) for c in color)
+
+    if shadow_color is None:
+        shadow_color = (0, 0, 0)
+    else:
+        shadow_color = tuple(int(c) for c in shadow_color)
+
+    if outline_color is None:
+        outline_color = (0, 0, 0)
+    else:
+        outline_color = tuple(int(c) for c in outline_color)
+
+    # Determine bounding box
+    if x is None:
+        x = 0
+        if width is None:
+            width = w
+    if y is None:
+        y = 0
+        if height is None:
+            height = h
+
+    x, y = int(x), int(y)
+    box_width = int(width) if width else w - x
+    box_height = int(height) if height else h - y
+
+    # Get or create font
+    if font is None:
+        if fit:
+            font = prim_fit_font(text, box_width, box_height, font_name)
+        elif font_name:
+            try:
+                font = prim_make_font(font_name, int(font_size))
+            except:
+                font = _get_default_font(int(font_size))
+        else:
+            font = _get_default_font(int(font_size))
+
+    # Measure text
+    text_w, text_h = prim_text_size(text, font)
+
+    # Calculate position based on alignment
+    if align == "center":
+        draw_x = x + (box_width - text_w) // 2
+    elif align == "right":
+        draw_x = x + box_width - text_w
+    else:  # left
+        draw_x = x
+
+    if valign == "middle":
+        draw_y = y + (box_height - text_h) // 2
+    elif valign == "bottom":
+        draw_y = y + box_height - text_h
+    else:  # top
+        draw_y = y
+
+    # Create RGBA image for compositing with opacity
+    pil_img = Image.fromarray(img).convert('RGBA')
+
+    # Create text layer with transparency
+    text_layer = Image.new('RGBA', (w, h), (0, 0, 0, 0))
+    draw = ImageDraw.Draw(text_layer)
+
+    # Draw shadow first (if enabled)
+    if shadow:
+        shadow_x = draw_x + shadow_offset
+        shadow_y = draw_y + shadow_offset
+        shadow_rgba = shadow_color + (int(255 * opacity * 0.5),)
+        draw.text((shadow_x, shadow_y), text, fill=shadow_rgba, font=font)
+
+    # Draw outline (if enabled)
+    if outline:
+        outline_rgba = outline_color + (int(255 * opacity),)
+        ow = int(outline_width)
+        for dx in range(-ow, ow + 1):
+            for dy in range(-ow, ow + 1):
+                if dx != 0 or dy != 0:
+                    draw.text((draw_x + dx, draw_y + dy), text,
+                             fill=outline_rgba, font=font)
+
+    # Draw main text
+    text_rgba = color + (int(255 * opacity),)
+    draw.text((draw_x, draw_y), text, fill=text_rgba, font=font)
+
+    # Composite
+    result = Image.alpha_composite(pil_img, text_layer)
+    return np.array(result.convert('RGB'))
+
+
+def prim_text_box(img: np.ndarray, text: str,
+                  x: int, y: int, width: int, height: int,
+                  font=None, font_size: int = 24, font_name: str = None,
+                  color=None, opacity: float = 1.0,
+                  align: str = "center", valign: str = "middle",
+                  fit: bool = True,
+                  padding: int = 0,
+                  background=None, background_opacity: float = 0.5,
+                  **kwargs) -> np.ndarray:
+    """
+    Draw text fitted within a box, optionally with background.
+
+    (text-box frame "Title" 50 50 400 100)
+    (text-box frame "Subtitle" 50 160 400 50
+              :background (0 0 0) :background-opacity 0.7)
+
+    Convenience wrapper around text() for common box-with-text pattern.
+    """
+    x, y = int(x), int(y)
+    width, height = int(width), int(height)
+    padding = int(padding)
+
+    result = img.copy()
+
+    # Draw background if specified
+    if background is not None:
+        bg_color = tuple(int(c) for c in background)
+
+        # Create background with opacity
+        pil_img = Image.fromarray(result).convert('RGBA')
+        bg_layer = Image.new('RGBA', (pil_img.width, pil_img.height), (0, 0, 0, 0))
+        bg_draw = ImageDraw.Draw(bg_layer)
+        bg_rgba = bg_color + (int(255 * background_opacity),)
+        bg_draw.rectangle([x, y, x + width, y + height], fill=bg_rgba)
+        result = np.array(Image.alpha_composite(pil_img, bg_layer).convert('RGB'))
+
+    # Draw text within padded box
+    return prim_text(result, text,
+                     x=x + padding, y=y + padding,
+                     width=width - 2 * padding, height=height - 2 * padding,
+                     font=font, font_size=font_size, font_name=font_name,
+                     color=color, opacity=opacity,
+                     align=align, valign=valign, fit=fit,
+                     **kwargs)
+
+
+# =============================================================================
+# Legacy text functions (keep for compatibility)
+# =============================================================================
 
 def prim_draw_char(img, char, x, y, font_size=16, color=None):
-    """Draw a single character at (x, y)."""
-    if color is None:
-        color = [255, 255, 255]
-
-    pil_img = Image.fromarray(img)
-    draw = ImageDraw.Draw(pil_img)
-    font = _get_default_font(font_size)
-    draw.text((x, y), char, fill=tuple(color), font=font)
-    return np.array(pil_img)
+    """Draw a single character at (x, y). Legacy function."""
+    return prim_text(img, str(char), x=int(x), y=int(y),
+                    font_size=int(font_size), color=color)
 
 
 def prim_draw_text(img, text, x, y, font_size=16, color=None):
-    """Draw text string at (x, y)."""
+    """Draw text string at (x, y). Legacy function."""
+    return prim_text(img, str(text), x=int(x), y=int(y),
+                    font_size=int(font_size), color=color)
+
+
+# =============================================================================
+# Shape Drawing
+# =============================================================================
+
+def prim_fill_rect(img, x, y, w, h, color=None, opacity: float = 1.0):
+    """
+    Fill a rectangle with color.
+
+    (fill-rect frame 10 10 100 50 (255 0 0))
+    (fill-rect frame 10 10 100 50 (255 0 0) :opacity 0.5)
+    """
     if color is None:
         color = [255, 255, 255]
 
-    pil_img = Image.fromarray(img)
-    draw = ImageDraw.Draw(pil_img)
-    font = _get_default_font(font_size)
-    draw.text((x, y), text, fill=tuple(color), font=font)
-    return np.array(pil_img)
-
-
-def prim_fill_rect(img, x, y, w, h, color=None):
-    """Fill a rectangle with color."""
-    if color is None:
-        color = [255, 255, 255]
-
-    result = img.copy()
     x, y, w, h = int(x), int(y), int(w), int(h)
-    result[y:y+h, x:x+w] = color
-    return result
+
+    if opacity >= 1.0:
+        result = img.copy()
+        result[y:y+h, x:x+w] = color
+        return result
+
+    # With opacity, use alpha compositing
+    pil_img = Image.fromarray(img).convert('RGBA')
+    layer = Image.new('RGBA', (pil_img.width, pil_img.height), (0, 0, 0, 0))
+    draw = ImageDraw.Draw(layer)
+    fill_rgba = tuple(int(c) for c in color) + (int(255 * opacity),)
+    draw.rectangle([x, y, x + w, y + h], fill=fill_rgba)
+    result = Image.alpha_composite(pil_img, layer)
+    return np.array(result.convert('RGB'))
 
 
-def prim_draw_rect(img, x, y, w, h, color=None, thickness=1):
+def prim_draw_rect(img, x, y, w, h, color=None, thickness=1, opacity: float = 1.0):
     """Draw rectangle outline."""
     if color is None:
         color = [255, 255, 255]
 
-    result = img.copy()
-    cv2.rectangle(result, (int(x), int(y)), (int(x+w), int(y+h)),
-                  tuple(color), thickness)
-    return result
+    if opacity >= 1.0:
+        result = img.copy()
+        cv2.rectangle(result, (int(x), int(y)), (int(x+w), int(y+h)),
+                      tuple(int(c) for c in color), int(thickness))
+        return result
+
+    # With opacity
+    pil_img = Image.fromarray(img).convert('RGBA')
+    layer = Image.new('RGBA', (pil_img.width, pil_img.height), (0, 0, 0, 0))
+    draw = ImageDraw.Draw(layer)
+    outline_rgba = tuple(int(c) for c in color) + (int(255 * opacity),)
+    draw.rectangle([int(x), int(y), int(x+w), int(y+h)],
+                   outline=outline_rgba, width=int(thickness))
+    result = Image.alpha_composite(pil_img, layer)
+    return np.array(result.convert('RGB'))
 
 
-def prim_draw_line(img, x1, y1, x2, y2, color=None, thickness=1):
+def prim_draw_line(img, x1, y1, x2, y2, color=None, thickness=1, opacity: float = 1.0):
     """Draw a line from (x1, y1) to (x2, y2)."""
     if color is None:
         color = [255, 255, 255]
 
-    result = img.copy()
-    cv2.line(result, (int(x1), int(y1)), (int(x2), int(y2)),
-             tuple(color), thickness)
-    return result
+    if opacity >= 1.0:
+        result = img.copy()
+        cv2.line(result, (int(x1), int(y1)), (int(x2), int(y2)),
+                 tuple(int(c) for c in color), int(thickness))
+        return result
+
+    # With opacity
+    pil_img = Image.fromarray(img).convert('RGBA')
+    layer = Image.new('RGBA', (pil_img.width, pil_img.height), (0, 0, 0, 0))
+    draw = ImageDraw.Draw(layer)
+    line_rgba = tuple(int(c) for c in color) + (int(255 * opacity),)
+    draw.line([(int(x1), int(y1)), (int(x2), int(y2))],
+              fill=line_rgba, width=int(thickness))
+    result = Image.alpha_composite(pil_img, layer)
+    return np.array(result.convert('RGB'))
 
 
-def prim_draw_circle(img, cx, cy, radius, color=None, thickness=1, fill=False):
+def prim_draw_circle(img, cx, cy, radius, color=None, thickness=1,
+                     fill=False, opacity: float = 1.0):
     """Draw a circle."""
     if color is None:
         color = [255, 255, 255]
 
-    result = img.copy()
-    t = -1 if fill else thickness
-    cv2.circle(result, (int(cx), int(cy)), int(radius), tuple(color), t)
-    return result
+    if opacity >= 1.0:
+        result = img.copy()
+        t = -1 if fill else int(thickness)
+        cv2.circle(result, (int(cx), int(cy)), int(radius),
+                   tuple(int(c) for c in color), t)
+        return result
+
+    # With opacity
+    pil_img = Image.fromarray(img).convert('RGBA')
+    layer = Image.new('RGBA', (pil_img.width, pil_img.height), (0, 0, 0, 0))
+    draw = ImageDraw.Draw(layer)
+    cx, cy, r = int(cx), int(cy), int(radius)
+    rgba = tuple(int(c) for c in color) + (int(255 * opacity),)
+
+    if fill:
+        draw.ellipse([cx - r, cy - r, cx + r, cy + r], fill=rgba)
+    else:
+        draw.ellipse([cx - r, cy - r, cx + r, cy + r],
+                     outline=rgba, width=int(thickness))
+
+    result = Image.alpha_composite(pil_img, layer)
+    return np.array(result.convert('RGB'))
 
 
-def prim_draw_ellipse(img, cx, cy, rx, ry, angle=0, color=None, thickness=1, fill=False):
+def prim_draw_ellipse(img, cx, cy, rx, ry, angle=0, color=None,
+                      thickness=1, fill=False, opacity: float = 1.0):
     """Draw an ellipse."""
     if color is None:
         color = [255, 255, 255]
 
-    result = img.copy()
-    t = -1 if fill else thickness
-    cv2.ellipse(result, (int(cx), int(cy)), (int(rx), int(ry)),
-                angle, 0, 360, tuple(color), t)
-    return result
+    if opacity >= 1.0:
+        result = img.copy()
+        t = -1 if fill else int(thickness)
+        cv2.ellipse(result, (int(cx), int(cy)), (int(rx), int(ry)),
+                    float(angle), 0, 360, tuple(int(c) for c in color), t)
+        return result
+
+    # With opacity (note: PIL doesn't support rotated ellipses easily)
+    # Fall back to cv2 on a separate layer
+    layer = np.zeros((img.shape[0], img.shape[1], 4), dtype=np.uint8)
+    t = -1 if fill else int(thickness)
+    rgba = tuple(int(c) for c in color) + (int(255 * opacity),)
+    cv2.ellipse(layer, (int(cx), int(cy)), (int(rx), int(ry)),
+                float(angle), 0, 360, rgba, t)
+
+    pil_img = Image.fromarray(img).convert('RGBA')
+    pil_layer = Image.fromarray(layer)
+    result = Image.alpha_composite(pil_img, pil_layer)
+    return np.array(result.convert('RGB'))
 
 
-def prim_draw_polygon(img, points, color=None, thickness=1, fill=False):
+def prim_draw_polygon(img, points, color=None, thickness=1,
+                      fill=False, opacity: float = 1.0):
     """Draw a polygon from list of [x, y] points."""
     if color is None:
         color = [255, 255, 255]
 
-    result = img.copy()
-    pts = np.array(points, dtype=np.int32).reshape((-1, 1, 2))
+    if opacity >= 1.0:
+        result = img.copy()
+        pts = np.array(points, dtype=np.int32).reshape((-1, 1, 2))
+        if fill:
+            cv2.fillPoly(result, [pts], tuple(int(c) for c in color))
+        else:
+            cv2.polylines(result, [pts], True,
+                         tuple(int(c) for c in color), int(thickness))
+        return result
+
+    # With opacity
+    pil_img = Image.fromarray(img).convert('RGBA')
+    layer = Image.new('RGBA', (pil_img.width, pil_img.height), (0, 0, 0, 0))
+    draw = ImageDraw.Draw(layer)
+
+    pts_flat = [(int(p[0]), int(p[1])) for p in points]
+    rgba = tuple(int(c) for c in color) + (int(255 * opacity),)
 
     if fill:
-        cv2.fillPoly(result, [pts], tuple(color))
+        draw.polygon(pts_flat, fill=rgba)
     else:
-        cv2.polylines(result, [pts], True, tuple(color), thickness)
+        draw.polygon(pts_flat, outline=rgba, width=int(thickness))
 
-    return result
+    result = Image.alpha_composite(pil_img, layer)
+    return np.array(result.convert('RGB'))
 
 
+# =============================================================================
+# PRIMITIVES Export
+# =============================================================================
+
 PRIMITIVES = {
-    # Text
+    # Font management
+    'make-font': prim_make_font,
+    'list-fonts': prim_list_fonts,
+    'font-size': prim_font_size,
+
+    # Text measurement
+    'text-size': prim_text_size,
+    'text-metrics': prim_text_metrics,
+    'fit-text-size': prim_fit_text_size,
+    'fit-font': prim_fit_font,
+
+    # Text drawing
+    'text': prim_text,
+    'text-box': prim_text_box,
+
+    # Legacy text (compatibility)
     'draw-char': prim_draw_char,
     'draw-text': prim_draw_text,
 

sexp_effects/primitive_libs/streaming.py

@@ -8,12 +8,18 @@ GPU Acceleration:
 - Set STREAMING_GPU_PERSIST=1 to output CuPy arrays (frames stay on GPU)
 - Hardware video decoding (NVDEC) is used when available
 - Dramatically improves performance on GPU nodes
+
+Async Prefetching:
+- Set STREAMING_PREFETCH=1 to enable background frame prefetching
+- Decodes upcoming frames while current frame is being processed
 """
 
 import os
 import numpy as np
 import subprocess
 import json
+import threading
+from collections import deque
 from pathlib import Path
 
 # Try to import CuPy for GPU acceleration
@@ -28,6 +34,10 @@ except ImportError:
 # Disabled by default until all primitives support GPU frames
 GPU_PERSIST = os.environ.get("STREAMING_GPU_PERSIST", "0") == "1" and CUPY_AVAILABLE
 
+# Async prefetch mode - decode frames in background thread
+PREFETCH_ENABLED = os.environ.get("STREAMING_PREFETCH", "1") == "1"
+PREFETCH_BUFFER_SIZE = int(os.environ.get("STREAMING_PREFETCH_SIZE", "10"))
+
 # Check for hardware decode support (cached)
 _HWDEC_AVAILABLE = None
 
@@ -283,6 +293,122 @@ class VideoSource:
             self._proc = None
 
 
+class PrefetchingVideoSource:
+    """
+    Video source with background prefetching for improved performance.
+
+    Wraps VideoSource and adds a background thread that pre-decodes
+    upcoming frames while the main thread processes the current frame.
+    """
+
+    def __init__(self, path: str, fps: float = 30, buffer_size: int = None):
+        self._source = VideoSource(path, fps)
+        self._buffer_size = buffer_size or PREFETCH_BUFFER_SIZE
+        self._buffer = {}  # time -> frame
+        self._buffer_lock = threading.Lock()
+        self._prefetch_time = 0.0
+        self._frame_time = 1.0 / fps
+        self._stop_event = threading.Event()
+        self._request_event = threading.Event()
+        self._target_time = 0.0
+
+        # Start prefetch thread
+        self._thread = threading.Thread(target=self._prefetch_loop, daemon=True)
+        self._thread.start()
+
+        import sys
+        print(f"PrefetchingVideoSource: {path} buffer_size={self._buffer_size}", file=sys.stderr)
+
+    def _prefetch_loop(self):
+        """Background thread that pre-reads frames."""
+        while not self._stop_event.is_set():
+            # Wait for work or timeout
+            self._request_event.wait(timeout=0.01)
+            self._request_event.clear()
+
+            if self._stop_event.is_set():
+                break
+
+            # Prefetch frames ahead of target time
+            target = self._target_time
+            with self._buffer_lock:
+                # Clean old frames (more than 1 second behind)
+                old_times = [t for t in self._buffer.keys() if t < target - 1.0]
+                for t in old_times:
+                    del self._buffer[t]
+
+                # Count how many frames we have buffered ahead
+                buffered_ahead = sum(1 for t in self._buffer.keys() if t >= target)
+
+            # Prefetch if buffer not full
+            if buffered_ahead < self._buffer_size:
+                # Find next time to prefetch
+                prefetch_t = target
+                with self._buffer_lock:
+                    existing_times = set(self._buffer.keys())
+                for _ in range(self._buffer_size):
+                    if prefetch_t not in existing_times:
+                        break
+                    prefetch_t += self._frame_time
+
+                # Read the frame (this is the slow part)
+                try:
+                    frame = self._source.read_at(prefetch_t)
+                    with self._buffer_lock:
+                        self._buffer[prefetch_t] = frame
+                except Exception as e:
+                    import sys
+                    print(f"Prefetch error at t={prefetch_t}: {e}", file=sys.stderr)
+
+    def read_at(self, t: float) -> np.ndarray:
+        """Read frame at specific time, using prefetch buffer if available."""
+        self._target_time = t
+        self._request_event.set()  # Wake up prefetch thread
+
+        # Round to frame time for buffer lookup
+        t_key = round(t / self._frame_time) * self._frame_time
+
+        # Check buffer first
+        with self._buffer_lock:
+            if t_key in self._buffer:
+                return self._buffer[t_key]
+            # Also check for close matches (within half frame time)
+            for buf_t, frame in self._buffer.items():
+                if abs(buf_t - t) < self._frame_time * 0.5:
+                    return frame
+
+        # Not in buffer - read directly (blocking)
+        frame = self._source.read_at(t)
+
+        # Store in buffer
+        with self._buffer_lock:
+            self._buffer[t_key] = frame
+
+        return frame
+
+    def read(self) -> np.ndarray:
+        """Read frame (uses last cached or t=0)."""
+        return self.read_at(0)
+
+    def skip(self):
+        """No-op for seek-based reading."""
+        pass
+
+    @property
+    def size(self):
+        return self._source.size
+
+    @property
+    def path(self):
+        return self._source.path
+
+    def close(self):
+        self._stop_event.set()
+        self._request_event.set()  # Wake up thread to exit
+        self._thread.join(timeout=1.0)
+        self._source.close()
+
+
 class AudioAnalyzer:
     """Audio analyzer for energy and beat detection."""
 
@@ -394,7 +520,12 @@ class AudioAnalyzer:
 # === Primitives ===
 
 def prim_make_video_source(path: str, fps: float = 30):
-    """Create a video source from a file path."""
+    """Create a video source from a file path.
+
+    Uses PrefetchingVideoSource if STREAMING_PREFETCH=1 (default).
+    """
+    if PREFETCH_ENABLED:
+        return PrefetchingVideoSource(path, fps)
     return VideoSource(path, fps)
 
 

sexp_effects/primitives.py

@@ -797,31 +797,63 @@ def prim_tan(x: float) -> float:
     return math.tan(x)
 
 
-def prim_atan2(y: float, x: float) -> float:
+def prim_atan2(y, x):
+    if hasattr(y, '_data'):  # Xector
+        from sexp_effects.primitive_libs.xector import Xector
+        return Xector(np.arctan2(y._data, x._data if hasattr(x, '_data') else x), y._shape)
     return math.atan2(y, x)
 
 
-def prim_sqrt(x: float) -> float:
+def prim_sqrt(x):
+    if hasattr(x, '_data'):  # Xector
+        from sexp_effects.primitive_libs.xector import Xector
+        return Xector(np.sqrt(np.maximum(0, x._data)), x._shape)
+    if isinstance(x, np.ndarray):
+        return np.sqrt(np.maximum(0, x))
     return math.sqrt(max(0, x))
 
 
-def prim_pow(x: float, y: float) -> float:
+def prim_pow(x, y):
+    if hasattr(x, '_data'):  # Xector
+        from sexp_effects.primitive_libs.xector import Xector
+        y_data = y._data if hasattr(y, '_data') else y
+        return Xector(np.power(x._data, y_data), x._shape)
     return math.pow(x, y)
 
 
-def prim_abs(x: float) -> float:
+def prim_abs(x):
+    if hasattr(x, '_data'):  # Xector
+        from sexp_effects.primitive_libs.xector import Xector
+        return Xector(np.abs(x._data), x._shape)
+    if isinstance(x, np.ndarray):
+        return np.abs(x)
     return abs(x)
 
 
-def prim_floor(x: float) -> int:
+def prim_floor(x):
+    if hasattr(x, '_data'):  # Xector
+        from sexp_effects.primitive_libs.xector import Xector
+        return Xector(np.floor(x._data), x._shape)
+    if isinstance(x, np.ndarray):
+        return np.floor(x)
     return int(math.floor(x))
 
 
-def prim_ceil(x: float) -> int:
+def prim_ceil(x):
+    if hasattr(x, '_data'):  # Xector
+        from sexp_effects.primitive_libs.xector import Xector
+        return Xector(np.ceil(x._data), x._shape)
+    if isinstance(x, np.ndarray):
+        return np.ceil(x)
     return int(math.ceil(x))
 
 
-def prim_round(x: float) -> int:
+def prim_round(x):
+    if hasattr(x, '_data'):  # Xector
+        from sexp_effects.primitive_libs.xector import Xector
+        return Xector(np.round(x._data), x._shape)
+    if isinstance(x, np.ndarray):
+        return np.round(x)
     return int(round(x))
 
 

streaming/jax_typography.py

@@ -0,0 +1,860 @@
+"""
+JAX Typography Primitives
+
+Two approaches for text rendering, both compile to JAX/GPU:
+
+## 1. TextStrip - Pixel-perfect static text
+   Pre-render entire strings at compile time using PIL.
+   Perfect sub-pixel anti-aliasing, exact match with PIL.
+   Use for: static titles, labels, any text without per-character effects.
+
+   S-expression:
+       (let ((strip (render-text-strip "Hello World" 48)))
+         (place-text-strip frame strip x y :color white))
+
+## 2. Glyph-by-glyph - Dynamic text effects
+   Individual glyph placement for wave, arc, audio-reactive effects.
+   Each character can have independent position, color, opacity.
+   Note: slight anti-aliasing differences vs PIL due to integer positioning.
+
+   S-expression:
+       ; Wave text - y oscillates with character index
+       (let ((glyphs (text-glyphs "Wavy" 48)))
+         (first
+           (fold glyphs (list frame 0)
+             (lambda (acc g)
+               (let ((frm (first acc))
+                     (cursor (second acc))
+                     (i (length acc)))  ; approximate index
+                 (list
+                   (place-glyph frm (glyph-image g)
+                     (+ x cursor)
+                     (+ y (* amplitude (sin (* i frequency))))
+                     (glyph-bearing-x g) (glyph-bearing-y g)
+                     white 1.0)
+                   (+ cursor (glyph-advance g))))))))
+
+       ; Audio-reactive spacing
+       (let ((glyphs (text-glyphs "Bass" 48))
+             (bass (audio-band 0 200)))
+         (first
+           (fold glyphs (list frame 0)
+             (lambda (acc g)
+               (let ((frm (first acc))
+                     (cursor (second acc)))
+                 (list
+                   (place-glyph frm (glyph-image g)
+                     (+ x cursor) y
+                     (glyph-bearing-x g) (glyph-bearing-y g)
+                     white 1.0)
+                   (+ cursor (glyph-advance g) (* bass 20))))))))
+
+Kerning support:
+    ; With kerning adjustment
+    (+ cursor (glyph-advance g) (glyph-kerning g next-g font-size))
+"""
+
+import numpy as np
+import jax.numpy as jnp
+from typing import Tuple, Dict, Any
+from dataclasses import dataclass
+
+
+# =============================================================================
+# Glyph Data (computed at compile time)
+# =============================================================================
+
+@dataclass
+class GlyphData:
+    """Glyph data computed at compile time.
+
+    Attributes:
+        char: The character
+        image: RGBA image as numpy array (H, W, 4) - converted to JAX at runtime
+        advance: Horizontal advance (distance to next glyph origin)
+        bearing_x: Left side bearing (x offset from origin to first pixel)
+        bearing_y: Top bearing (y offset from baseline to top of glyph)
+        width: Image width
+        height: Image height
+    """
+    char: str
+    image: np.ndarray  # (H, W, 4) RGBA uint8
+    advance: float
+    bearing_x: float
+    bearing_y: float
+    width: int
+    height: int
+
+
+# Font cache: (font_name, font_size) -> {char: GlyphData}
+_GLYPH_CACHE: Dict[Tuple, Dict[str, GlyphData]] = {}
+
+# Font metrics cache: (font_name, font_size) -> (ascent, descent)
+_METRICS_CACHE: Dict[Tuple, Tuple[float, float]] = {}
+
+# Kerning cache: (font_name, font_size) -> {(char1, char2): adjustment}
+# Kerning adjustment is added to advance: new_advance = advance + kerning
+# Typically negative (characters move closer together)
+_KERNING_CACHE: Dict[Tuple, Dict[Tuple[str, str], float]] = {}
+
+
+def _load_font(font_name: str = None, font_size: int = 32):
+    """Load a font. Called at compile time."""
+    from PIL import ImageFont
+
+    candidates = [
+        font_name,
+        '/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf',
+        '/usr/share/fonts/truetype/liberation/LiberationSans-Regular.ttf',
+        '/usr/share/fonts/truetype/freefont/FreeSans.ttf',
+    ]
+
+    for path in candidates:
+        if path is None:
+            continue
+        try:
+            return ImageFont.truetype(path, font_size)
+        except (IOError, OSError):
+            continue
+
+    return ImageFont.load_default()
+
+
+def _get_glyph_cache(font_name: str = None, font_size: int = 32) -> Dict[str, GlyphData]:
+    """Get or create glyph cache for a font. Called at compile time."""
+    cache_key = (font_name, font_size)
+
+    if cache_key in _GLYPH_CACHE:
+        return _GLYPH_CACHE[cache_key]
+
+    from PIL import Image, ImageDraw
+
+    font = _load_font(font_name, font_size)
+    ascent, descent = font.getmetrics()
+    _METRICS_CACHE[cache_key] = (ascent, descent)
+
+    glyphs = {}
+    charset = ''.join(chr(i) for i in range(32, 127))
+
+    temp_img = Image.new('RGBA', (200, 200), (0, 0, 0, 0))
+    temp_draw = ImageDraw.Draw(temp_img)
+
+    for char in charset:
+        # Get metrics
+        bbox = temp_draw.textbbox((0, 0), char, font=font)
+        advance = font.getlength(char)
+
+        x_min, y_min, x_max, y_max = bbox
+
+        # Create glyph image with padding
+        padding = 2
+        img_w = max(int(x_max - x_min) + padding * 2, 1)
+        img_h = max(int(y_max - y_min) + padding * 2, 1)
+
+        glyph_img = Image.new('RGBA', (img_w, img_h), (0, 0, 0, 0))
+        glyph_draw = ImageDraw.Draw(glyph_img)
+
+        # Draw at position accounting for bbox offset
+        draw_x = padding - x_min
+        draw_y = padding - y_min
+        glyph_draw.text((draw_x, draw_y), char, fill=(255, 255, 255, 255), font=font)
+
+        glyphs[char] = GlyphData(
+            char=char,
+            image=np.array(glyph_img, dtype=np.uint8),
+            advance=float(advance),
+            bearing_x=float(x_min),
+            bearing_y=float(-y_min),  # Distance from baseline to top
+            width=img_w,
+            height=img_h,
+        )
+
+    _GLYPH_CACHE[cache_key] = glyphs
+    return glyphs
+
+
+def _get_kerning_cache(font_name: str = None, font_size: int = 32) -> Dict[Tuple[str, str], float]:
+    """Get or create kerning cache for a font. Called at compile time.
+
+    Kerning is computed as:
+        kerning(a, b) = getlength(a + b) - getlength(a) - getlength(b)
+
+    This gives the adjustment needed when placing 'b' after 'a'.
+    Typically negative (characters move closer together).
+    """
+    cache_key = (font_name, font_size)
+
+    if cache_key in _KERNING_CACHE:
+        return _KERNING_CACHE[cache_key]
+
+    font = _load_font(font_name, font_size)
+    kerning = {}
+
+    # Compute kerning for all printable ASCII pairs
+    charset = ''.join(chr(i) for i in range(32, 127))
+
+    # Pre-compute individual character lengths
+    char_lengths = {c: font.getlength(c) for c in charset}
+
+    # Compute kerning for each pair
+    for c1 in charset:
+        for c2 in charset:
+            pair_length = font.getlength(c1 + c2)
+            individual_sum = char_lengths[c1] + char_lengths[c2]
+            kern = pair_length - individual_sum
+
+            # Only store non-zero kerning to save memory
+            if abs(kern) > 0.01:
+                kerning[(c1, c2)] = kern
+
+    _KERNING_CACHE[cache_key] = kerning
+    return kerning
+
+
+def get_kerning(char1: str, char2: str, font_name: str = None, font_size: int = 32) -> float:
+    """Get kerning adjustment between two characters. Compile-time.
+
+    Returns the adjustment to add to char1's advance when char2 follows.
+    Typically negative (characters move closer).
+
+    Usage in S-expression:
+        (+ (glyph-advance g1) (kerning g1 g2))
+    """
+    kerning_cache = _get_kerning_cache(font_name, font_size)
+    return kerning_cache.get((char1, char2), 0.0)
+
+
+@dataclass
+class TextStrip:
+    """Pre-rendered text strip with proper sub-pixel anti-aliasing.
+
+    Rendered at compile time using PIL for exact matching.
+    At runtime, just composite onto frame at integer positions.
+
+    Attributes:
+        text: The original text
+        image: RGBA image as numpy array (H, W, 4)
+        width: Strip width
+        height: Strip height
+        baseline_y: Y position of baseline within the strip
+        bearing_x: Left side bearing of first character
+        anchor_x: X offset for anchor point (0 for left, width/2 for center, width for right)
+        anchor_y: Y offset for anchor point (depends on anchor type)
+        stroke_width: Stroke width used when rendering
+    """
+    text: str
+    image: np.ndarray
+    width: int
+    height: int
+    baseline_y: int
+    bearing_x: float
+    anchor_x: float = 0.0
+    anchor_y: float = 0.0
+    stroke_width: int = 0
+
+
+# Text strip cache: cache_key -> TextStrip
+_TEXT_STRIP_CACHE: Dict[Tuple, TextStrip] = {}
+
+
+def render_text_strip(
+    text: str,
+    font_name: str = None,
+    font_size: int = 32,
+    stroke_width: int = 0,
+    stroke_fill: tuple = None,
+    anchor: str = "la",  # left-ascender (PIL default is "la")
+    multiline: bool = False,
+    line_spacing: int = 4,
+    align: str = "left",
+) -> TextStrip:
+    """Render text to a strip at compile time. Perfect sub-pixel anti-aliasing.
+
+    Args:
+        text: Text to render
+        font_name: Path to font file (None for default)
+        font_size: Font size in pixels
+        stroke_width: Outline width in pixels (0 for no outline)
+        stroke_fill: Outline color as (R,G,B) or (R,G,B,A), default black
+        anchor: PIL anchor code - first char: h=left, m=middle, r=right
+                                  second char: a=ascender, t=top, m=middle, s=baseline, d=descender
+        multiline: If True, handle newlines in text
+        line_spacing: Extra pixels between lines (for multiline)
+        align: 'left', 'center', 'right' (for multiline)
+
+    Returns:
+        TextStrip with pre-rendered text
+    """
+    # Build cache key from all parameters
+    cache_key = (text, font_name, font_size, stroke_width, stroke_fill, anchor, multiline, line_spacing, align)
+    if cache_key in _TEXT_STRIP_CACHE:
+        return _TEXT_STRIP_CACHE[cache_key]
+
+    from PIL import Image, ImageDraw
+
+    font = _load_font(font_name, font_size)
+    ascent, descent = font.getmetrics()
+
+    # Default stroke fill to black
+    if stroke_fill is None:
+        stroke_fill = (0, 0, 0, 255)
+    elif len(stroke_fill) == 3:
+        stroke_fill = (*stroke_fill, 255)
+
+    # Get text bbox (accounting for stroke)
+    temp = Image.new('RGBA', (1, 1))
+    temp_draw = ImageDraw.Draw(temp)
+
+    if multiline:
+        bbox = temp_draw.multiline_textbbox((0, 0), text, font=font, spacing=line_spacing,
+                                             stroke_width=stroke_width)
+    else:
+        bbox = temp_draw.textbbox((0, 0), text, font=font, stroke_width=stroke_width)
+
+    # bbox is (left, top, right, bottom) relative to origin
+    x_min, y_min, x_max, y_max = bbox
+
+    # Create image with padding (extra for stroke)
+    padding = 2 + stroke_width
+    img_width = max(int(x_max - x_min) + padding * 2, 1)
+    img_height = max(int(y_max - y_min) + padding * 2, 1)
+
+    # Create RGBA image
+    img = Image.new('RGBA', (img_width, img_height), (0, 0, 0, 0))
+    draw = ImageDraw.Draw(img)
+
+    # Draw text at position that puts it in the image
+    draw_x = padding - x_min
+    draw_y = padding - y_min
+
+    if multiline:
+        draw.multiline_text((draw_x, draw_y), text, fill=(255, 255, 255, 255), font=font,
+                           spacing=line_spacing, align=align,
+                           stroke_width=stroke_width, stroke_fill=stroke_fill)
+    else:
+        draw.text((draw_x, draw_y), text, fill=(255, 255, 255, 255), font=font,
+                  stroke_width=stroke_width, stroke_fill=stroke_fill)
+
+    # Baseline is at y=0 in text coordinates, which is at draw_y in image
+    baseline_y = draw_y
+
+    # Convert to numpy for pixel analysis
+    img_array = np.array(img, dtype=np.uint8)
+
+    # Calculate anchor offsets
+    # For 'm' (middle) anchors, compute from actual rendered pixels for pixel-perfect matching
+    h_anchor = anchor[0] if len(anchor) > 0 else 'l'
+    v_anchor = anchor[1] if len(anchor) > 1 else 'a'
+
+    # Find actual pixel bounds (for middle anchor calculations)
+    alpha = img_array[:, :, 3]
+    nonzero_cols = np.where(alpha.max(axis=0) > 0)[0]
+    nonzero_rows = np.where(alpha.max(axis=1) > 0)[0]
+
+    if len(nonzero_cols) > 0:
+        pixel_x_min = nonzero_cols.min()
+        pixel_x_max = nonzero_cols.max()
+        pixel_x_center = (pixel_x_min + pixel_x_max) / 2.0
+    else:
+        pixel_x_center = img_width / 2.0
+
+    if len(nonzero_rows) > 0:
+        pixel_y_min = nonzero_rows.min()
+        pixel_y_max = nonzero_rows.max()
+        pixel_y_center = (pixel_y_min + pixel_y_max) / 2.0
+    else:
+        pixel_y_center = img_height / 2.0
+
+    # Horizontal offset
+    text_width = x_max - x_min
+    if h_anchor == 'l':  # left edge of text
+        anchor_x = float(draw_x)
+    elif h_anchor == 'm':  # middle - use actual pixel center for perfect matching
+        anchor_x = pixel_x_center
+    elif h_anchor == 'r':  # right edge of text
+        anchor_x = float(draw_x + text_width)
+    else:
+        anchor_x = float(draw_x)
+
+    # Vertical offset
+    # PIL anchor positions are based on font metrics (ascent/descent):
+    # - 'a' (ascender): at the ascender line → draw_y in strip
+    # - 't' (top): at top of text bounding box → padding in strip
+    # - 'm' (middle): center of em-square = (ascent + descent) / 2 below ascender
+    # - 's' (baseline): at baseline = ascent below ascender
+    # - 'd' (descender): at descender line = ascent + descent below ascender
+
+    if v_anchor == 'a':  # ascender
+        anchor_y = float(draw_y)
+    elif v_anchor == 't':  # top of bbox
+        anchor_y = float(padding)
+    elif v_anchor == 'm':  # middle (center of em-square, per PIL's calculation)
+        anchor_y = float(draw_y + (ascent + descent) / 2.0)
+    elif v_anchor == 's':  # baseline
+        anchor_y = float(draw_y + ascent)
+    elif v_anchor == 'd':  # descender
+        anchor_y = float(draw_y + ascent + descent)
+    else:
+        anchor_y = float(draw_y)  # default to ascender
+
+    strip = TextStrip(
+        text=text,
+        image=img_array,
+        width=img_width,
+        height=img_height,
+        baseline_y=baseline_y,
+        bearing_x=float(x_min),
+        anchor_x=anchor_x,
+        anchor_y=anchor_y,
+        stroke_width=stroke_width,
+    )
+
+    _TEXT_STRIP_CACHE[cache_key] = strip
+    return strip
+
+
+# =============================================================================
+# Compile-time functions (called during S-expression compilation)
+# =============================================================================
+
+def get_glyph(char: str, font_name: str = None, font_size: int = 32) -> GlyphData:
+    """Get glyph data for a single character. Compile-time."""
+    cache = _get_glyph_cache(font_name, font_size)
+    return cache.get(char, cache.get(' '))
+
+
+def get_glyphs(text: str, font_name: str = None, font_size: int = 32) -> list:
+    """Get glyph data for a string. Compile-time."""
+    cache = _get_glyph_cache(font_name, font_size)
+    space = cache.get(' ')
+    return [cache.get(c, space) for c in text]
+
+
+def get_font_ascent(font_name: str = None, font_size: int = 32) -> float:
+    """Get font ascent. Compile-time."""
+    _get_glyph_cache(font_name, font_size)  # Ensure cache exists
+    return _METRICS_CACHE[(font_name, font_size)][0]
+
+
+def get_font_descent(font_name: str = None, font_size: int = 32) -> float:
+    """Get font descent. Compile-time."""
+    _get_glyph_cache(font_name, font_size)
+    return _METRICS_CACHE[(font_name, font_size)][1]
+
+
+# =============================================================================
+# JAX Runtime Primitives
+# =============================================================================
+
+def place_glyph_jax(
+    frame: jnp.ndarray,
+    glyph_image: jnp.ndarray,  # (H, W, 4) RGBA
+    x: float,
+    y: float,
+    bearing_x: float,
+    bearing_y: float,
+    color: jnp.ndarray,  # (3,) RGB 0-255
+    opacity: float = 1.0,
+) -> jnp.ndarray:
+    """
+    Place a glyph onto a frame. This is the core JAX primitive.
+
+    All positioning math can use traced values (x, y from audio, time, etc.)
+    The glyph_image is static (determined at compile time).
+
+    Args:
+        frame: (H, W, 3) RGB frame
+        glyph_image: (gh, gw, 4) RGBA glyph (pre-converted to JAX array)
+        x: X position of glyph origin (baseline point)
+        y: Y position of baseline
+        bearing_x: Left side bearing
+        bearing_y: Top bearing (from baseline to top)
+        color: RGB color array
+        opacity: Opacity 0-1
+
+    Returns:
+        Frame with glyph composited
+    """
+    h, w = frame.shape[:2]
+    gh, gw = glyph_image.shape[:2]
+
+    # Calculate destination position
+    # bearing_x: how far right of origin the glyph starts (can be negative)
+    # bearing_y: how far up from baseline the glyph extends
+    padding = 2  # Must match padding used in glyph creation
+    dst_x = x + bearing_x - padding
+    dst_y = y - bearing_y - padding
+
+    # Extract glyph RGB and alpha
+    glyph_rgb = glyph_image[:, :, :3].astype(jnp.float32) / 255.0
+    # Match PIL's integer alpha math: (coverage * int(opacity*255) + 127) // 255
+    opacity_int = jnp.round(opacity * 255)
+    glyph_a_raw = glyph_image[:, :, 3:4].astype(jnp.float32)
+    glyph_alpha = jnp.floor(glyph_a_raw * opacity_int / 255.0 + 0.5) / 255.0
+
+    # Apply color tint (glyph is white, multiply by color)
+    color_normalized = color.astype(jnp.float32) / 255.0
+    tinted = glyph_rgb * color_normalized
+
+    from jax.lax import dynamic_update_slice
+
+    # Use padded buffer to avoid XLA's dynamic_update_slice clamping
+    buf_h = h + 2 * gh
+    buf_w = w + 2 * gw
+    rgb_buf = jnp.zeros((buf_h, buf_w, 3), dtype=jnp.float32)
+    alpha_buf = jnp.zeros((buf_h, buf_w, 1), dtype=jnp.float32)
+
+    dst_x_int = dst_x.astype(jnp.int32)
+    dst_y_int = dst_y.astype(jnp.int32)
+    place_y = jnp.maximum(dst_y_int + gh, 0).astype(jnp.int32)
+    place_x = jnp.maximum(dst_x_int + gw, 0).astype(jnp.int32)
+
+    rgb_buf = dynamic_update_slice(rgb_buf, tinted, (place_y, place_x, 0))
+    alpha_buf = dynamic_update_slice(alpha_buf, glyph_alpha, (place_y, place_x, 0))
+
+    rgb_layer = rgb_buf[gh:gh + h, gw:gw + w, :]
+    alpha_layer = alpha_buf[gh:gh + h, gw:gw + w, :]
+
+    # Alpha composite using PIL-compatible integer arithmetic
+    src_int = jnp.floor(rgb_layer * 255 + 0.5).astype(jnp.int32)
+    alpha_int = jnp.floor(alpha_layer * 255 + 0.5).astype(jnp.int32)
+    dst_int = frame.astype(jnp.int32)
+    result = (src_int * alpha_int + dst_int * (255 - alpha_int) + 127) // 255
+
+    return jnp.clip(result, 0, 255).astype(jnp.uint8)
+
+
+def place_text_strip_jax(
+    frame: jnp.ndarray,
+    strip_image: jnp.ndarray,  # (H, W, 4) RGBA
+    x: float,
+    y: float,
+    baseline_y: int,
+    bearing_x: float,
+    color: jnp.ndarray,
+    opacity: float = 1.0,
+    anchor_x: float = 0.0,
+    anchor_y: float = 0.0,
+    stroke_width: int = 0,
+) -> jnp.ndarray:
+    """
+    Place a pre-rendered text strip onto a frame.
+
+    The strip was rendered at compile time with proper sub-pixel anti-aliasing.
+    This just composites it at the specified position.
+
+    Args:
+        frame: (H, W, 3) RGB frame
+        strip_image: (sh, sw, 4) RGBA text strip
+        x: X position for anchor point
+        y: Y position for anchor point
+        baseline_y: Y position of baseline within the strip
+        bearing_x: Left side bearing
+        color: RGB color
+        opacity: Opacity 0-1
+        anchor_x: X offset of anchor point within strip
+        anchor_y: Y offset of anchor point within strip
+        stroke_width: Stroke width used when rendering (affects padding)
+
+    Returns:
+        Frame with text composited
+    """
+    h, w = frame.shape[:2]
+    sh, sw = strip_image.shape[:2]
+
+    # Calculate destination position
+    # Anchor point (anchor_x, anchor_y) in strip should be at (x, y) in frame
+    # anchor_x/anchor_y already account for the anchor position within the strip
+    # Use floor(x + 0.5) for consistent rounding (jnp.round uses banker's rounding)
+    dst_x = jnp.floor(x - anchor_x + 0.5).astype(jnp.int32)
+    dst_y = jnp.floor(y - anchor_y + 0.5).astype(jnp.int32)
+
+    # Extract strip RGB and alpha
+    strip_rgb = strip_image[:, :, :3].astype(jnp.float32) / 255.0
+    # Match PIL's integer alpha math: (coverage * int(opacity*255) + 127) // 255
+    # Use jnp.round (banker's rounding) to match Python's round() used by PIL
+    opacity_int = jnp.round(opacity * 255)
+    strip_a_raw = strip_image[:, :, 3:4].astype(jnp.float32)
+    strip_alpha = jnp.floor(strip_a_raw * opacity_int / 255.0 + 0.5) / 255.0
+
+    # Apply color tint
+    color_normalized = color.astype(jnp.float32) / 255.0
+    tinted = strip_rgb * color_normalized
+
+    from jax.lax import dynamic_update_slice
+
+    # Use a padded buffer to avoid XLA's dynamic_update_slice clamping behavior.
+    # XLA clamps indices so the update fits, which silently shifts the strip.
+    # By placing into a buffer padded by strip dimensions, then extracting the
+    # frame-sized region, we get correct clipping for both overflow and underflow.
+    buf_h = h + 2 * sh
+    buf_w = w + 2 * sw
+    rgb_buf = jnp.zeros((buf_h, buf_w, 3), dtype=jnp.float32)
+    alpha_buf = jnp.zeros((buf_h, buf_w, 1), dtype=jnp.float32)
+
+    # Offset by (sh, sw) so dst=0 maps to (sh, sw) in buffer
+    place_y = jnp.maximum(dst_y + sh, 0).astype(jnp.int32)
+    place_x = jnp.maximum(dst_x + sw, 0).astype(jnp.int32)
+
+    rgb_buf = dynamic_update_slice(rgb_buf, tinted, (place_y, place_x, 0))
+    alpha_buf = dynamic_update_slice(alpha_buf, strip_alpha, (place_y, place_x, 0))
+
+    # Extract frame-sized region (sh, sw are compile-time constants from strip shape)
+    rgb_layer = rgb_buf[sh:sh + h, sw:sw + w, :]
+    alpha_layer = alpha_buf[sh:sh + h, sw:sw + w, :]
+
+    # Alpha composite using PIL-compatible integer arithmetic:
+    # result = (src * alpha + dst * (255 - alpha) + 127) // 255
+    src_int = jnp.floor(rgb_layer * 255 + 0.5).astype(jnp.int32)
+    alpha_int = jnp.floor(alpha_layer * 255 + 0.5).astype(jnp.int32)
+    dst_int = frame.astype(jnp.int32)
+    result = (src_int * alpha_int + dst_int * (255 - alpha_int) + 127) // 255
+
+    return jnp.clip(result, 0, 255).astype(jnp.uint8)
+
+
+def place_glyph_simple(
+    frame: jnp.ndarray,
+    glyph: GlyphData,
+    x: float,
+    y: float,
+    color: tuple = (255, 255, 255),
+    opacity: float = 1.0,
+) -> jnp.ndarray:
+    """
+    Convenience wrapper that takes GlyphData directly.
+    Converts glyph image to JAX array.
+
+    For S-expression use, prefer place_glyph_jax with pre-converted arrays.
+    """
+    glyph_jax = jnp.asarray(glyph.image)
+    color_jax = jnp.array(color, dtype=jnp.float32)
+
+    return place_glyph_jax(
+        frame, glyph_jax, x, y,
+        glyph.bearing_x, glyph.bearing_y,
+        color_jax, opacity
+    )
+
+
+# =============================================================================
+# S-Expression Primitive Bindings
+# =============================================================================
+
+def bind_typography_primitives(env: dict) -> dict:
+    """
+    Add typography primitives to an S-expression environment.
+
+    Primitives added:
+        (text-glyphs text font-size) -> list of glyph data
+        (glyph-image g) -> JAX array (H, W, 4)
+        (glyph-advance g) -> float
+        (glyph-bearing-x g) -> float
+        (glyph-bearing-y g) -> float
+        (glyph-width g) -> int
+        (glyph-height g) -> int
+        (font-ascent font-size) -> float
+        (font-descent font-size) -> float
+        (place-glyph frame glyph-img x y bearing-x bearing-y color opacity) -> frame
+    """
+
+    def prim_text_glyphs(text, font_size=32, font_name=None):
+        """Get list of glyph data for text. Compile-time."""
+        return get_glyphs(str(text), font_name, int(font_size))
+
+    def prim_glyph_image(glyph):
+        """Get glyph image as JAX array."""
+        return jnp.asarray(glyph.image)
+
+    def prim_glyph_advance(glyph):
+        """Get glyph advance width."""
+        return glyph.advance
+
+    def prim_glyph_bearing_x(glyph):
+        """Get glyph left side bearing."""
+        return glyph.bearing_x
+
+    def prim_glyph_bearing_y(glyph):
+        """Get glyph top bearing."""
+        return glyph.bearing_y
+
+    def prim_glyph_width(glyph):
+        """Get glyph image width."""
+        return glyph.width
+
+    def prim_glyph_height(glyph):
+        """Get glyph image height."""
+        return glyph.height
+
+    def prim_font_ascent(font_size=32, font_name=None):
+        """Get font ascent."""
+        return get_font_ascent(font_name, int(font_size))
+
+    def prim_font_descent(font_size=32, font_name=None):
+        """Get font descent."""
+        return get_font_descent(font_name, int(font_size))
+
+    def prim_place_glyph(frame, glyph_img, x, y, bearing_x, bearing_y,
+                         color=(255, 255, 255), opacity=1.0):
+        """Place glyph on frame. Runtime JAX operation."""
+        color_arr = jnp.array(color, dtype=jnp.float32)
+        return place_glyph_jax(frame, glyph_img, x, y, bearing_x, bearing_y,
+                               color_arr, opacity)
+
+    def prim_glyph_kerning(glyph1, glyph2, font_size=32, font_name=None):
+        """Get kerning adjustment between two glyphs. Compile-time.
+
+        Returns adjustment to add to glyph1's advance when glyph2 follows.
+        Typically negative (characters move closer).
+
+        Usage: (+ (glyph-advance g) (glyph-kerning g next-g font-size))
+        """
+        return get_kerning(glyph1.char, glyph2.char, font_name, int(font_size))
+
+    def prim_char_kerning(char1, char2, font_size=32, font_name=None):
+        """Get kerning adjustment between two characters. Compile-time."""
+        return get_kerning(str(char1), str(char2), font_name, int(font_size))
+
+    # TextStrip primitives for pre-rendered text with proper anti-aliasing
+    def prim_render_text_strip(text, font_size=32, font_name=None):
+        """Render text to a strip at compile time. Perfect anti-aliasing."""
+        return render_text_strip(str(text), font_name, int(font_size))
+
+    def prim_render_text_strip_styled(
+        text, font_size=32, font_name=None,
+        stroke_width=0, stroke_fill=None,
+        anchor="la", multiline=False, line_spacing=4, align="left"
+    ):
+        """Render styled text to a strip. Supports stroke, anchors, multiline.
+
+        Args:
+            text: Text to render
+            font_size: Size in pixels
+            font_name: Path to font file
+            stroke_width: Outline width (0 = no outline)
+            stroke_fill: Outline color as (R,G,B) or (R,G,B,A)
+            anchor: 2-char anchor code (e.g., "mm" for center, "la" for left-ascender)
+            multiline: If True, handle newlines
+            line_spacing: Extra pixels between lines
+            align: "left", "center", "right" for multiline
+        """
+        return render_text_strip(
+            str(text), font_name, int(font_size),
+            stroke_width=int(stroke_width),
+            stroke_fill=stroke_fill,
+            anchor=str(anchor),
+            multiline=bool(multiline),
+            line_spacing=int(line_spacing),
+            align=str(align),
+        )
+
+    def prim_text_strip_image(strip):
+        """Get text strip image as JAX array."""
+        return jnp.asarray(strip.image)
+
+    def prim_text_strip_width(strip):
+        """Get text strip width."""
+        return strip.width
+
+    def prim_text_strip_height(strip):
+        """Get text strip height."""
+        return strip.height
+
+    def prim_text_strip_baseline_y(strip):
+        """Get text strip baseline Y position."""
+        return strip.baseline_y
+
+    def prim_text_strip_bearing_x(strip):
+        """Get text strip left bearing."""
+        return strip.bearing_x
+
+    def prim_text_strip_anchor_x(strip):
+        """Get text strip anchor X offset."""
+        return strip.anchor_x
+
+    def prim_text_strip_anchor_y(strip):
+        """Get text strip anchor Y offset."""
+        return strip.anchor_y
+
+    def prim_place_text_strip(frame, strip, x, y, color=(255, 255, 255), opacity=1.0):
+        """Place pre-rendered text strip on frame. Runtime JAX operation."""
+        strip_img = jnp.asarray(strip.image)
+        color_arr = jnp.array(color, dtype=jnp.float32)
+        return place_text_strip_jax(
+            frame, strip_img, x, y,
+            strip.baseline_y, strip.bearing_x,
+            color_arr, opacity,
+            anchor_x=strip.anchor_x, anchor_y=strip.anchor_y,
+            stroke_width=strip.stroke_width
+        )
+
+    # Add to environment
+    env.update({
+        # Glyph-by-glyph primitives (for wave, arc, audio-reactive effects)
+        'text-glyphs': prim_text_glyphs,
+        'glyph-image': prim_glyph_image,
+        'glyph-advance': prim_glyph_advance,
+        'glyph-bearing-x': prim_glyph_bearing_x,
+        'glyph-bearing-y': prim_glyph_bearing_y,
+        'glyph-width': prim_glyph_width,
+        'glyph-height': prim_glyph_height,
+        'glyph-kerning': prim_glyph_kerning,
+        'char-kerning': prim_char_kerning,
+        'font-ascent': prim_font_ascent,
+        'font-descent': prim_font_descent,
+        'place-glyph': prim_place_glyph,
+        # TextStrip primitives (for pixel-perfect static text)
+        'render-text-strip': prim_render_text_strip,
+        'render-text-strip-styled': prim_render_text_strip_styled,
+        'text-strip-image': prim_text_strip_image,
+        'text-strip-width': prim_text_strip_width,
+        'text-strip-height': prim_text_strip_height,
+        'text-strip-baseline-y': prim_text_strip_baseline_y,
+        'text-strip-bearing-x': prim_text_strip_bearing_x,
+        'text-strip-anchor-x': prim_text_strip_anchor_x,
+        'text-strip-anchor-y': prim_text_strip_anchor_y,
+        'place-text-strip': prim_place_text_strip,
+    })
+
+    return env
+
+
+# =============================================================================
+# Example: Render text using primitives (for testing)
+# =============================================================================
+
+def render_text_primitives(
+    frame: jnp.ndarray,
+    text: str,
+    x: float,
+    y: float,
+    font_size: int = 32,
+    color: tuple = (255, 255, 255),
+    opacity: float = 1.0,
+    use_kerning: bool = True,
+) -> jnp.ndarray:
+    """
+    Render text using the primitives.
+    This is what an S-expression would compile to.
+
+    Args:
+        use_kerning: If True, apply kerning adjustments between characters
+    """
+    glyphs = get_glyphs(text, None, font_size)
+    color_arr = jnp.array(color, dtype=jnp.float32)
+
+    cursor = x
+    for i, g in enumerate(glyphs):
+        glyph_jax = jnp.asarray(g.image)
+        frame = place_glyph_jax(
+            frame, glyph_jax, cursor, y,
+            g.bearing_x, g.bearing_y,
+            color_arr, opacity
+        )
+        # Advance cursor with optional kerning
+        advance = g.advance
+        if use_kerning and i + 1 < len(glyphs):
+            advance += get_kerning(g.char, glyphs[i + 1].char, None, font_size)
+        cursor = cursor + advance
+
+    return frame

streaming/stream_sexp_generic.py

@@ -21,6 +21,7 @@ Context (ctx) is passed explicitly to frame evaluation:
 """
 
 import sys
+import os
 import time
 import json
 import hashlib
@@ -62,6 +63,38 @@ class Context:
     fps: float = 30.0
 
 
+class DeferredEffectChain:
+    """
+    Represents a chain of JAX effects that haven't been executed yet.
+
+    Allows effects to be accumulated through let bindings and fused
+    into a single JIT-compiled function when the result is needed.
+    """
+    __slots__ = ('effects', 'params_list', 'base_frame', 't', 'frame_num')
+
+    def __init__(self, effects: list, params_list: list, base_frame, t: float, frame_num: int):
+        self.effects = effects  # List of effect names, innermost first
+        self.params_list = params_list  # List of param dicts, matching effects
+        self.base_frame = base_frame  # The actual frame array at the start
+        self.t = t
+        self.frame_num = frame_num
+
+    def extend(self, effect_name: str, params: dict) -> 'DeferredEffectChain':
+        """Add another effect to the chain (outermost)."""
+        return DeferredEffectChain(
+            self.effects + [effect_name],
+            self.params_list + [params],
+            self.base_frame,
+            self.t,
+            self.frame_num
+        )
+
+    @property
+    def shape(self):
+        """Allow shape check without forcing execution."""
+        return self.base_frame.shape if hasattr(self.base_frame, 'shape') else None
+
+
 class StreamInterpreter:
     """
     Fully generic streaming sexp interpreter.
@@ -98,6 +131,9 @@ class StreamInterpreter:
         self.use_jax = use_jax
         self.jax_effects: Dict[str, Callable] = {}  # Cache of JAX-compiled effects
         self.jax_effect_paths: Dict[str, Path] = {}  # Track source paths for effects
+        self.jax_fused_chains: Dict[str, Callable] = {}  # Cache of fused effect chains
+        self.jax_batched_chains: Dict[str, Callable] = {}  # Cache of vmapped chains
+        self.jax_batch_size: int = int(os.environ.get("JAX_BATCH_SIZE", "30"))  # Configurable via env
         if use_jax:
             if _init_jax():
                 print("JAX acceleration enabled", file=sys.stderr)
@@ -238,6 +274,8 @@ class StreamInterpreter:
         """Load primitives from a Python library file.
 
         Prefers GPU-accelerated versions (*_gpu.py) when available.
+        Uses cached modules from sys.modules to ensure consistent state
+        (e.g., same RNG instance for all interpreters).
         """
         import importlib.util
 
@@ -264,9 +302,26 @@ class StreamInterpreter:
         if not lib_path:
             raise FileNotFoundError(f"Primitive library '{lib_name}' not found. Searched paths: {lib_paths}")
 
-        spec = importlib.util.spec_from_file_location(actual_lib_name, lib_path)
-        module = importlib.util.module_from_spec(spec)
-        spec.loader.exec_module(module)
+        # Use cached module if already imported to preserve state (e.g., RNG)
+        # This is critical for deterministic random number sequences
+        # Check multiple possible module keys (standard import paths and our cache)
+        possible_keys = [
+            f"sexp_effects.primitive_libs.{actual_lib_name}",
+            f"sexp_primitives.{actual_lib_name}",
+        ]
+
+        module = None
+        for key in possible_keys:
+            if key in sys.modules:
+                module = sys.modules[key]
+                break
+
+        if module is None:
+            spec = importlib.util.spec_from_file_location(actual_lib_name, lib_path)
+            module = importlib.util.module_from_spec(spec)
+            spec.loader.exec_module(module)
+            # Cache for future use under our key
+            sys.modules[f"sexp_primitives.{actual_lib_name}"] = module
 
         # Check if this is a GPU-accelerated module
         is_gpu = actual_lib_name.endswith('_gpu')
@@ -452,30 +507,353 @@ class StreamInterpreter:
 
         try:
             jax_fn = self.jax_effects[name]
-            # Ensure frame is numpy array
+            # Handle GPU frames (CuPy) - need to move to CPU for CPU JAX
+            # JAX handles numpy and JAX arrays natively, no conversion needed
             if hasattr(frame, 'cpu'):
                 frame = frame.cpu
-            elif hasattr(frame, 'get'):
-                frame = frame.get()
+            elif hasattr(frame, 'get') and hasattr(frame, '__cuda_array_interface__'):
+                frame = frame.get()  # CuPy array -> numpy
 
             # Get seed from config for deterministic random
             seed = self.config.get('seed', 42)
 
             # Call JAX function with parameters
-            result = jax_fn(frame, t=t, frame_num=frame_num, seed=seed, **params)
-
-            # Convert result back to numpy if needed
-            if hasattr(result, 'block_until_ready'):
-                result.block_until_ready()  # Ensure computation is complete
-            if hasattr(result, '__array__'):
-                result = np.asarray(result)
-
-            return result
+            # Return JAX array directly - don't block or convert per-effect
+            # Conversion to numpy happens once at frame write time
+            return jax_fn(frame, t=t, frame_num=frame_num, seed=seed, **params)
         except Exception as e:
             # Fall back to interpreter on error
             print(f"JAX effect {name} error, falling back: {e}", file=sys.stderr)
             return None
 
+    def _is_jax_effect_expr(self, expr) -> bool:
+        """Check if an expression is a JAX-compiled effect call."""
+        if not isinstance(expr, list) or not expr:
+            return False
+        head = expr[0]
+        if not isinstance(head, Symbol):
+            return False
+        return head.name in self.jax_effects
+
+    def _extract_effect_chain(self, expr, env) -> Optional[Tuple[list, list, Any]]:
+        """
+        Extract a chain of JAX effects from an expression.
+
+        Returns: (effect_names, params_list, base_frame_expr) or None if not a chain.
+        effect_names and params_list are in execution order (innermost first).
+
+        For (effect1 (effect2 frame :p1 v1) :p2 v2):
+        Returns: (['effect2', 'effect1'], [params2, params1], frame_expr)
+        """
+        if not self._is_jax_effect_expr(expr):
+            return None
+
+        chain = []
+        params_list = []
+        current = expr
+
+        while self._is_jax_effect_expr(current):
+            head = current[0]
+            effect_name = head.name
+            args = current[1:]
+
+            # Extract params for this effect
+            effect = self.effects[effect_name]
+            effect_params = {}
+            for pname, pdef in effect['params'].items():
+                effect_params[pname] = pdef.get('default', 0)
+
+            # Find the frame argument (first positional) and other params
+            frame_arg = None
+            i = 0
+            while i < len(args):
+                if isinstance(args[i], Keyword):
+                    pname = args[i].name
+                    if pname in effect['params'] and i + 1 < len(args):
+                        effect_params[pname] = self._eval(args[i + 1], env)
+                    i += 2
+                else:
+                    if frame_arg is None:
+                        frame_arg = args[i]  # First positional is frame
+                    i += 1
+
+            chain.append(effect_name)
+            params_list.append(effect_params)
+
+            if frame_arg is None:
+                return None  # No frame argument found
+
+            # Check if frame_arg is another effect call
+            if self._is_jax_effect_expr(frame_arg):
+                current = frame_arg
+            else:
+                # End of chain - frame_arg is the base frame
+                # Reverse to get innermost-first execution order
+                chain.reverse()
+                params_list.reverse()
+                return (chain, params_list, frame_arg)
+
+        return None
+
+    def _get_chain_key(self, effect_names: list, params_list: list) -> str:
+        """Generate a cache key for an effect chain.
+
+        Includes static param values in the key since they affect compilation.
+        """
+        parts = []
+        for name, params in zip(effect_names, params_list):
+            param_parts = []
+            for pname in sorted(params.keys()):
+                pval = params[pname]
+                # Include static values in key (strings, bools)
+                if isinstance(pval, (str, bool)):
+                    param_parts.append(f"{pname}={pval}")
+                else:
+                    param_parts.append(pname)
+            parts.append(f"{name}:{','.join(param_parts)}")
+        return '|'.join(parts)
+
+    def _compile_effect_chain(self, effect_names: list, params_list: list) -> Optional[Callable]:
+        """
+        Compile a chain of effects into a single fused JAX function.
+
+        Args:
+            effect_names: List of effect names in order [innermost, ..., outermost]
+            params_list: List of param dicts for each effect (used to detect static types)
+
+        Returns:
+            JIT-compiled function: (frame, t, frame_num, seed, **all_params) -> frame
+        """
+        if not _JAX_AVAILABLE:
+            return None
+
+        try:
+            import jax
+
+            # Get the individual JAX functions
+            jax_fns = [self.jax_effects[name] for name in effect_names]
+
+            # Pre-extract param names and identify static params from actual values
+            effect_param_names = []
+            static_params = ['seed']  # seed is always static
+            for i, (name, params) in enumerate(zip(effect_names, params_list)):
+                param_names = list(params.keys())
+                effect_param_names.append(param_names)
+                # Check actual values to identify static types
+                for pname, pval in params.items():
+                    if isinstance(pval, (str, bool)):
+                        static_params.append(f"_p{i}_{pname}")
+
+            def fused_fn(frame, t, frame_num, seed, **kwargs):
+                result = frame
+                for i, (jax_fn, param_names) in enumerate(zip(jax_fns, effect_param_names)):
+                    # Extract params for this effect from kwargs
+                    effect_kwargs = {}
+                    for pname in param_names:
+                        key = f"_p{i}_{pname}"
+                        if key in kwargs:
+                            effect_kwargs[pname] = kwargs[key]
+                    result = jax_fn(result, t=t, frame_num=frame_num, seed=seed, **effect_kwargs)
+                return result
+
+            # JIT with static params (seed + any string/bool params)
+            return jax.jit(fused_fn, static_argnames=static_params)
+        except Exception as e:
+            print(f"Failed to compile effect chain {effect_names}: {e}", file=sys.stderr)
+            return None
+
+    def _apply_effect_chain(self, effect_names: list, params_list: list, frame, t: float, frame_num: int):
+        """Apply a chain of effects, using fused compilation if available."""
+        chain_key = self._get_chain_key(effect_names, params_list)
+
+        # Try to get or compile fused chain
+        if chain_key not in self.jax_fused_chains:
+            fused_fn = self._compile_effect_chain(effect_names, params_list)
+            self.jax_fused_chains[chain_key] = fused_fn
+            if fused_fn:
+                print(f"  [JAX fused chain: {' -> '.join(effect_names)}]", file=sys.stderr)
+
+        fused_fn = self.jax_fused_chains.get(chain_key)
+
+        if fused_fn is not None:
+            # Build kwargs with prefixed param names
+            kwargs = {}
+            for i, params in enumerate(params_list):
+                for pname, pval in params.items():
+                    kwargs[f"_p{i}_{pname}"] = pval
+
+            seed = self.config.get('seed', 42)
+
+            # Handle GPU frames
+            if hasattr(frame, 'cpu'):
+                frame = frame.cpu
+            elif hasattr(frame, 'get') and hasattr(frame, '__cuda_array_interface__'):
+                frame = frame.get()
+
+            try:
+                return fused_fn(frame, t=t, frame_num=frame_num, seed=seed, **kwargs)
+            except Exception as e:
+                print(f"Fused chain error: {e}", file=sys.stderr)
+
+        # Fall back to sequential application
+        result = frame
+        for name, params in zip(effect_names, params_list):
+            result = self._apply_jax_effect(name, result, params, t, frame_num)
+            if result is None:
+                return None
+        return result
+
+    def _force_deferred(self, deferred: DeferredEffectChain):
+        """Execute a deferred effect chain and return the actual array."""
+        if len(deferred.effects) == 0:
+            return deferred.base_frame
+
+        return self._apply_effect_chain(
+            deferred.effects,
+            deferred.params_list,
+            deferred.base_frame,
+            deferred.t,
+            deferred.frame_num
+        )
+
+    def _maybe_force(self, value):
+        """Force a deferred chain if needed, otherwise return as-is."""
+        if isinstance(value, DeferredEffectChain):
+            return self._force_deferred(value)
+        return value
+
+    def _compile_batched_chain(self, effect_names: list, params_list: list) -> Optional[Callable]:
+        """
+        Compile a vmapped version of an effect chain for batch processing.
+
+        Args:
+            effect_names: List of effect names in order [innermost, ..., outermost]
+            params_list: List of param dicts (used to detect static types)
+
+        Returns:
+            Batched function: (frames, ts, frame_nums, seed, **batched_params) -> frames
+            Where frames is (N, H, W, 3), ts/frame_nums are (N,), params are (N,) or scalar
+        """
+        if not _JAX_AVAILABLE:
+            return None
+
+        try:
+            import jax
+            import jax.numpy as jnp
+
+            # Get the individual JAX functions
+            jax_fns = [self.jax_effects[name] for name in effect_names]
+
+            # Pre-extract param info
+            effect_param_names = []
+            static_params = set()
+            for i, (name, params) in enumerate(zip(effect_names, params_list)):
+                param_names = list(params.keys())
+                effect_param_names.append(param_names)
+                for pname, pval in params.items():
+                    if isinstance(pval, (str, bool)):
+                        static_params.add(f"_p{i}_{pname}")
+
+            # Single-frame function (will be vmapped)
+            def single_frame_fn(frame, t, frame_num, seed, **kwargs):
+                result = frame
+                for i, (jax_fn, param_names) in enumerate(zip(jax_fns, effect_param_names)):
+                    effect_kwargs = {}
+                    for pname in param_names:
+                        key = f"_p{i}_{pname}"
+                        if key in kwargs:
+                            effect_kwargs[pname] = kwargs[key]
+                    result = jax_fn(result, t=t, frame_num=frame_num, seed=seed, **effect_kwargs)
+                return result
+
+            # Return unbatched function - we'll vmap at call time with proper in_axes
+            return jax.jit(single_frame_fn, static_argnames=['seed'] + list(static_params))
+        except Exception as e:
+            print(f"Failed to compile batched chain {effect_names}: {e}", file=sys.stderr)
+            return None
+
+    def _apply_batched_chain(self, effect_names: list, params_list_batch: list,
+                             frames: list, ts: list, frame_nums: list) -> Optional[list]:
+        """
+        Apply an effect chain to a batch of frames using vmap.
+
+        Args:
+            effect_names: List of effect names
+            params_list_batch: List of params_list for each frame in batch
+            frames: List of input frames
+            ts: List of time values
+            frame_nums: List of frame numbers
+
+        Returns:
+            List of output frames, or None on failure
+        """
+        if not frames:
+            return []
+
+        # Use first frame's params for chain key (assume same structure)
+        chain_key = self._get_chain_key(effect_names, params_list_batch[0])
+        batch_key = f"batch:{chain_key}"
+
+        # Compile batched version if needed
+        if batch_key not in self.jax_batched_chains:
+            batched_fn = self._compile_batched_chain(effect_names, params_list_batch[0])
+            self.jax_batched_chains[batch_key] = batched_fn
+            if batched_fn:
+                print(f"  [JAX batched chain: {' -> '.join(effect_names)} x{len(frames)}]", file=sys.stderr)
+
+        batched_fn = self.jax_batched_chains.get(batch_key)
+
+        if batched_fn is not None:
+            try:
+                import jax
+                import jax.numpy as jnp
+
+                # Stack frames into batch array
+                frames_array = jnp.stack([f if not hasattr(f, 'get') else f.get() for f in frames])
+                ts_array = jnp.array(ts)
+                frame_nums_array = jnp.array(frame_nums)
+
+                # Build kwargs - all numeric params as arrays for vmap
+                kwargs = {}
+                static_kwargs = {}  # Non-vmapped (strings, bools)
+
+                for i, plist in enumerate(zip(*[p for p in params_list_batch])):
+                    for j, pname in enumerate(params_list_batch[0][i].keys()):
+                        key = f"_p{i}_{pname}"
+                        values = [p[pname] for p in [params_list_batch[b][i] for b in range(len(frames))]]
+
+                        first = values[0]
+                        if isinstance(first, (str, bool)):
+                            # Static params - not vmapped
+                            static_kwargs[key] = first
+                        elif isinstance(first, (int, float)):
+                            # Always batch numeric params for simplicity
+                            kwargs[key] = jnp.array(values)
+                        elif hasattr(first, 'shape'):
+                            kwargs[key] = jnp.stack(values)
+                        else:
+                            kwargs[key] = jnp.array(values)
+
+                seed = self.config.get('seed', 42)
+
+                # Create wrapper that unpacks the params dict
+                def single_call(frame, t, frame_num, params_dict):
+                    return batched_fn(frame, t, frame_num, seed, **params_dict, **static_kwargs)
+
+                # vmap over frame, t, frame_num, and the params dict (as pytree)
+                vmapped_fn = jax.vmap(single_call, in_axes=(0, 0, 0, 0))
+
+                # Stack kwargs into a dict of arrays (pytree with matching structure)
+                results = vmapped_fn(frames_array, ts_array, frame_nums_array, kwargs)
+
+                # Unstack results
+                return [results[i] for i in range(len(frames))]
+            except Exception as e:
+                print(f"Batched chain error: {e}", file=sys.stderr)
+
+        # Fall back to sequential
+        return None
+
     def _init(self):
         """Initialize from sexp - load primitives, effects, defs, scans."""
         # Set random seed for deterministic output
@@ -869,6 +1247,22 @@ class StreamInterpreter:
         # === Effects ===
 
         if op in self.effects:
+            # Try to detect and fuse effect chains for JAX acceleration
+            if self.use_jax and op in self.jax_effects:
+                chain_info = self._extract_effect_chain(expr, env)
+                if chain_info is not None:
+                    effect_names, params_list, base_frame_expr = chain_info
+                    # Only use chain if we have 2+ effects (worth fusing)
+                    if len(effect_names) >= 2:
+                        base_frame = self._eval(base_frame_expr, env)
+                        if base_frame is not None and hasattr(base_frame, 'shape'):
+                            t = env.get('t', 0.0)
+                            frame_num = env.get('frame-num', 0)
+                            result = self._apply_effect_chain(effect_names, params_list, base_frame, t, frame_num)
+                            if result is not None:
+                                return result
+                            # Fall through if chain application fails
+
             effect = self.effects[op]
             effect_env = dict(env)
 
@@ -895,17 +1289,28 @@ class StreamInterpreter:
                     positional_idx += 1
                     i += 1
 
-            # Try JAX-accelerated execution first
+            # Try JAX-accelerated execution with deferred chaining
             if self.use_jax and op in self.jax_effects and frame_val is not None:
                 # Build params dict for JAX (exclude 'frame')
-                jax_params = {k: v for k, v in effect_env.items()
+                jax_params = {k: self._maybe_force(v) for k, v in effect_env.items()
                              if k != 'frame' and k in effect['params']}
                 t = env.get('t', 0.0)
                 frame_num = env.get('frame-num', 0)
-                result = self._apply_jax_effect(op, frame_val, jax_params, t, frame_num)
-                if result is not None:
-                    return result
-                # Fall through to interpreter if JAX fails
+
+                # Check if input is a deferred chain - if so, extend it
+                if isinstance(frame_val, DeferredEffectChain):
+                    return frame_val.extend(op, jax_params)
+
+                # Check if input is a valid frame - create new deferred chain
+                if hasattr(frame_val, 'shape'):
+                    return DeferredEffectChain([op], [jax_params], frame_val, t, frame_num)
+
+                # Fall through to interpreter if not a valid frame
+
+            # Force any deferred frame before interpreter evaluation
+            if isinstance(frame_val, DeferredEffectChain):
+                frame_val = self._force_deferred(frame_val)
+                effect_env['frame'] = frame_val
 
             return self._eval(effect['body'], effect_env)
 
@@ -922,10 +1327,15 @@ class StreamInterpreter:
                 if isinstance(args[i], Keyword):
                     k = args[i].name
                     v = self._eval(args[i + 1], env) if i + 1 < len(args) else None
+                    # Force deferred chains before passing to primitives
+                    v = self._maybe_force(v)
                     kwargs[k] = self._maybe_to_numpy(v, for_gpu_primitive=is_gpu_prim)
                     i += 2
                 else:
-                    evaluated_args.append(self._maybe_to_numpy(self._eval(args[i], env), for_gpu_primitive=is_gpu_prim))
+                    val = self._eval(args[i], env)
+                    # Force deferred chains before passing to primitives
+                    val = self._maybe_force(val)
+                    evaluated_args.append(self._maybe_to_numpy(val, for_gpu_primitive=is_gpu_prim))
                     i += 1
             try:
                 if kwargs:
@@ -1152,6 +1562,61 @@ class StreamInterpreter:
             eval_times = []
             write_times = []
 
+            # Batch accumulation for JAX
+            batch_deferred = []  # Accumulated DeferredEffectChains
+            batch_times = []  # Corresponding time values
+            batch_start_frame = 0
+
+            def flush_batch():
+                """Execute accumulated batch and write results."""
+                nonlocal batch_deferred, batch_times
+                if not batch_deferred:
+                    return
+
+                t_flush = time.time()
+
+                # Check if all chains have same structure (can batch)
+                first = batch_deferred[0]
+                can_batch = (
+                    self.use_jax and
+                    len(batch_deferred) >= 2 and
+                    all(d.effects == first.effects for d in batch_deferred)
+                )
+
+                if can_batch:
+                    # Try batched execution
+                    frames = [d.base_frame for d in batch_deferred]
+                    ts = [d.t for d in batch_deferred]
+                    frame_nums = [d.frame_num for d in batch_deferred]
+                    params_batch = [d.params_list for d in batch_deferred]
+
+                    results = self._apply_batched_chain(
+                        first.effects, params_batch, frames, ts, frame_nums
+                    )
+
+                    if results is not None:
+                        # Write batched results
+                        for result, t in zip(results, batch_times):
+                            if hasattr(result, 'block_until_ready'):
+                                result.block_until_ready()
+                                result = np.asarray(result)
+                            out.write(result, t)
+                        batch_deferred = []
+                        batch_times = []
+                        return
+
+                # Fall back to sequential execution
+                for deferred, t in zip(batch_deferred, batch_times):
+                    result = self._force_deferred(deferred)
+                    if result is not None and hasattr(result, 'shape'):
+                        if hasattr(result, 'block_until_ready'):
+                            result.block_until_ready()
+                            result = np.asarray(result)
+                        out.write(result, t)
+
+                batch_deferred = []
+                batch_times = []
+
             for frame_num in range(start_frame, n_frames):
                 if not out.is_open:
                     break
@@ -1182,8 +1647,23 @@ class StreamInterpreter:
                 eval_times.append(time.time() - t1)
 
                 t2 = time.time()
-                if result is not None and hasattr(result, 'shape'):
-                    out.write(result, ctx.t)
+                if result is not None:
+                    if isinstance(result, DeferredEffectChain):
+                        # Accumulate for batching
+                        batch_deferred.append(result)
+                        batch_times.append(ctx.t)
+
+                        # Flush when batch is full
+                        if len(batch_deferred) >= self.jax_batch_size:
+                            flush_batch()
+                    else:
+                        # Not deferred - flush any pending batch first, then write
+                        flush_batch()
+                        if hasattr(result, 'shape'):
+                            if hasattr(result, 'block_until_ready'):
+                                result.block_until_ready()
+                                result = np.asarray(result)
+                            out.write(result, ctx.t)
                 write_times.append(time.time() - t2)
 
                 frame_elapsed = time.time() - frame_start
@@ -1219,6 +1699,9 @@ class StreamInterpreter:
                         except Exception as e:
                             print(f"Warning: progress callback failed: {e}", file=sys.stderr)
 
+            # Flush any remaining batch
+            flush_batch()
+
         finally:
             out.close()
             # Store output for access to properties like playlist_cid

test_funky_text.py

@@ -0,0 +1,542 @@
+#!/usr/bin/env python3
+"""
+Funky comparison tests: PIL vs TextStrip system.
+Tests colors, opacity, fonts, sizes, edge positions, clipping, overlaps, etc.
+"""
+
+import numpy as np
+import jax.numpy as jnp
+from PIL import Image, ImageDraw, ImageFont
+from streaming.jax_typography import (
+    render_text_strip, place_text_strip_jax, _load_font
+)
+
+FONTS = {
+    'sans': '/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf',
+    'bold': '/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf',
+    'serif': '/usr/share/fonts/truetype/dejavu/DejaVuSerif.ttf',
+    'serif_bold': '/usr/share/fonts/truetype/dejavu/DejaVuSerif-Bold.ttf',
+    'mono': '/usr/share/fonts/truetype/dejavu/DejaVuSansMono.ttf',
+    'mono_bold': '/usr/share/fonts/truetype/dejavu/DejaVuSansMono-Bold.ttf',
+    'narrow': '/usr/share/fonts/truetype/liberation/LiberationSansNarrow-Regular.ttf',
+    'italic': '/usr/share/fonts/truetype/freefont/FreeSansOblique.ttf',
+}
+
+
+def render_pil(text, x, y, font_path=None, font_size=36, frame_size=(400, 100),
+               fill=(255, 255, 255), bg=(0, 0, 0), opacity=1.0,
+               stroke_width=0, stroke_fill=None, anchor="la",
+               multiline=False, line_spacing=4, align="left"):
+    """Render with PIL directly, including color/opacity."""
+    frame = np.full((frame_size[1], frame_size[0], 3), bg, dtype=np.uint8)
+    # For opacity, render to RGBA then composite
+    txt_layer = Image.new('RGBA', frame_size, (0, 0, 0, 0))
+    draw = ImageDraw.Draw(txt_layer)
+    font = _load_font(font_path, font_size)
+
+    if stroke_fill is None:
+        stroke_fill = (0, 0, 0)
+
+    # PIL fill with alpha for opacity
+    alpha_int = int(round(opacity * 255))
+    fill_rgba = (*fill, alpha_int)
+    stroke_rgba = (*stroke_fill, alpha_int) if stroke_width > 0 else None
+
+    if multiline:
+        draw.multiline_text((x, y), text, fill=fill_rgba, font=font,
+                           stroke_width=stroke_width, stroke_fill=stroke_rgba,
+                           spacing=line_spacing, align=align, anchor=anchor)
+    else:
+        draw.text((x, y), text, fill=fill_rgba, font=font,
+                  stroke_width=stroke_width, stroke_fill=stroke_rgba, anchor=anchor)
+
+    # Composite onto background
+    bg_img = Image.fromarray(frame).convert('RGBA')
+    result = Image.alpha_composite(bg_img, txt_layer)
+    return np.array(result.convert('RGB'))
+
+
+def render_strip(text, x, y, font_path=None, font_size=36, frame_size=(400, 100),
+                 fill=(255, 255, 255), bg=(0, 0, 0), opacity=1.0,
+                 stroke_width=0, stroke_fill=None, anchor="la",
+                 multiline=False, line_spacing=4, align="left"):
+    """Render with TextStrip system."""
+    frame = jnp.full((frame_size[1], frame_size[0], 3), jnp.array(bg, dtype=jnp.uint8), dtype=jnp.uint8)
+
+    strip = render_text_strip(
+        text, font_path, font_size,
+        stroke_width=stroke_width, stroke_fill=stroke_fill,
+        anchor=anchor, multiline=multiline, line_spacing=line_spacing, align=align
+    )
+    strip_img = jnp.asarray(strip.image)
+    color = jnp.array(fill, dtype=jnp.float32)
+
+    result = place_text_strip_jax(
+        frame, strip_img, x, y,
+        strip.baseline_y, strip.bearing_x,
+        color, opacity,
+        anchor_x=strip.anchor_x, anchor_y=strip.anchor_y,
+        stroke_width=strip.stroke_width
+    )
+    return np.array(result)
+
+
+def compare(name, tolerance=0, **kwargs):
+    """Compare PIL and TextStrip rendering."""
+    pil = render_pil(**kwargs)
+    strip = render_strip(**kwargs)
+
+    diff = np.abs(pil.astype(np.int16) - strip.astype(np.int16))
+    max_diff = diff.max()
+    pixels_diff = (diff > 0).any(axis=2).sum()
+
+    if max_diff == 0:
+        print(f"  PASS: {name}")
+        return True
+
+    if tolerance > 0:
+        best_diff = diff.copy()
+        for dy in range(-tolerance, tolerance + 1):
+            for dx in range(-tolerance, tolerance + 1):
+                if dy == 0 and dx == 0:
+                    continue
+                shifted = np.roll(np.roll(strip, dy, axis=0), dx, axis=1)
+                sdiff = np.abs(pil.astype(np.int16) - shifted.astype(np.int16))
+                best_diff = np.minimum(best_diff, sdiff)
+        if best_diff.max() == 0:
+            print(f"  PASS: {name} (within {tolerance}px tolerance)")
+            return True
+
+    print(f"  FAIL: {name}")
+    print(f"        Max diff: {max_diff}, Pixels different: {pixels_diff}")
+    Image.fromarray(pil).save(f"/tmp/pil_{name}.png")
+    Image.fromarray(strip).save(f"/tmp/strip_{name}.png")
+    diff_vis = np.clip(diff * 10, 0, 255).astype(np.uint8)
+    Image.fromarray(diff_vis).save(f"/tmp/diff_{name}.png")
+    print(f"        Saved: /tmp/pil_{name}.png /tmp/strip_{name}.png /tmp/diff_{name}.png")
+    return False
+
+
+def test_colors():
+    """Test various text colors on various backgrounds."""
+    print("\n--- Colors ---")
+    results = []
+
+    # White on black (baseline)
+    results.append(compare("color_white_on_black",
+        text="Hello", x=20, y=30, fill=(255, 255, 255), bg=(0, 0, 0)))
+
+    # Red on black
+    results.append(compare("color_red",
+        text="Red Text", x=20, y=30, fill=(255, 0, 0), bg=(0, 0, 0)))
+
+    # Green on black
+    results.append(compare("color_green",
+        text="Green!", x=20, y=30, fill=(0, 255, 0), bg=(0, 0, 0)))
+
+    # Blue on black
+    results.append(compare("color_blue",
+        text="Blue sky", x=20, y=30, fill=(0, 100, 255), bg=(0, 0, 0)))
+
+    # Yellow on dark gray
+    results.append(compare("color_yellow_on_gray",
+        text="Yellow", x=20, y=30, fill=(255, 255, 0), bg=(40, 40, 40)))
+
+    # Magenta on white
+    results.append(compare("color_magenta_on_white",
+        text="Magenta", x=20, y=30, fill=(255, 0, 255), bg=(255, 255, 255)))
+
+    # Subtle: gray text on slightly lighter gray
+    results.append(compare("color_subtle_gray",
+        text="Subtle", x=20, y=30, fill=(128, 128, 128), bg=(64, 64, 64)))
+
+    # Orange on deep blue
+    results.append(compare("color_orange_on_blue",
+        text="Warm", x=20, y=30, fill=(255, 165, 0), bg=(0, 0, 80)))
+
+    return results
+
+
+def test_opacity():
+    """Test different opacity levels."""
+    print("\n--- Opacity ---")
+    results = []
+
+    results.append(compare("opacity_100",
+        text="Full", x=20, y=30, opacity=1.0))
+
+    results.append(compare("opacity_75",
+        text="75%", x=20, y=30, opacity=0.75))
+
+    results.append(compare("opacity_50",
+        text="Half", x=20, y=30, opacity=0.5))
+
+    results.append(compare("opacity_25",
+        text="Quarter", x=20, y=30, opacity=0.25))
+
+    results.append(compare("opacity_10",
+        text="Ghost", x=20, y=30, opacity=0.1))
+
+    # Opacity on colored background
+    results.append(compare("opacity_on_colored_bg",
+        text="Overlay", x=20, y=30, fill=(255, 255, 255), bg=(100, 0, 0),
+        opacity=0.5))
+
+    # Color + opacity combo
+    results.append(compare("opacity_red_on_green",
+        text="Blend", x=20, y=30, fill=(255, 0, 0), bg=(0, 100, 0),
+        opacity=0.6))
+
+    return results
+
+
+def test_fonts():
+    """Test different fonts and sizes."""
+    print("\n--- Fonts & Sizes ---")
+    results = []
+
+    for label, path in FONTS.items():
+        results.append(compare(f"font_{label}",
+            text="Quick Fox", x=20, y=30, font_path=path, font_size=28,
+            frame_size=(300, 80)))
+
+    # Tiny text
+    results.append(compare("size_tiny",
+        text="Tiny text at 12px", x=10, y=15, font_size=12,
+        frame_size=(200, 40)))
+
+    # Big text
+    results.append(compare("size_big",
+        text="BIG", x=20, y=10, font_size=72,
+        frame_size=(300, 100)))
+
+    # Huge text
+    results.append(compare("size_huge",
+        text="XL", x=10, y=10, font_size=120,
+        frame_size=(300, 160)))
+
+    return results
+
+
+def test_anchors():
+    """Test all anchor combinations."""
+    print("\n--- Anchors ---")
+    results = []
+
+    # All horizontal x vertical combos
+    for h in ['l', 'm', 'r']:
+        for v in ['a', 'm', 's', 'd']:
+            anchor = h + v
+            results.append(compare(f"anchor_{anchor}",
+                text="Anchor", x=200, y=50, anchor=anchor,
+                frame_size=(400, 100)))
+
+    return results
+
+
+def test_strokes():
+    """Test various stroke widths and colors."""
+    print("\n--- Strokes ---")
+    results = []
+
+    for sw in [1, 2, 3, 4, 6, 8]:
+        results.append(compare(f"stroke_w{sw}",
+            text="Stroke", x=30, y=20, font_size=40,
+            stroke_width=sw, stroke_fill=(0, 0, 0),
+            frame_size=(300, 80)))
+
+    # Colored strokes
+    results.append(compare("stroke_red",
+        text="Red outline", x=20, y=20, font_size=36,
+        stroke_width=3, stroke_fill=(255, 0, 0),
+        frame_size=(350, 80)))
+
+    results.append(compare("stroke_white_on_black",
+        text="Glow", x=20, y=20, font_size=40,
+        fill=(255, 255, 255), stroke_width=4, stroke_fill=(0, 0, 255),
+        frame_size=(250, 80)))
+
+    # Stroke with bold font
+    results.append(compare("stroke_bold",
+        text="Bold+Stroke", x=20, y=20,
+        font_path=FONTS['bold'], font_size=36,
+        stroke_width=3, stroke_fill=(0, 0, 0),
+        frame_size=(400, 80)))
+
+    # Stroke + colored text on colored bg
+    results.append(compare("stroke_colored_on_bg",
+        text="Party", x=20, y=20, font_size=48,
+        fill=(255, 255, 0), bg=(50, 0, 80),
+        stroke_width=3, stroke_fill=(255, 0, 0),
+        frame_size=(300, 80)))
+
+    return results
+
+
+def test_edge_clipping():
+    """Test text at frame edges - clipping behavior."""
+    print("\n--- Edge Clipping ---")
+    results = []
+
+    # Text at very left edge
+    results.append(compare("clip_left_edge",
+        text="LEFT", x=0, y=30, frame_size=(200, 80)))
+
+    # Text partially off right edge
+    results.append(compare("clip_right_edge",
+        text="RIGHT SIDE", x=150, y=30, frame_size=(200, 80)))
+
+    # Text at top edge
+    results.append(compare("clip_top",
+        text="TOP", x=20, y=0, frame_size=(200, 80)))
+
+    # Text at bottom edge - partially clipped
+    results.append(compare("clip_bottom",
+        text="BOTTOM", x=20, y=55, font_size=40,
+        frame_size=(200, 80)))
+
+    # Large text overflowing all sides from center
+    results.append(compare("clip_overflow_center",
+        text="HUGE", x=75, y=40, font_size=100, anchor="mm",
+        frame_size=(150, 80)))
+
+    # Corner placement
+    results.append(compare("clip_corner_br",
+        text="Corner", x=350, y=70, font_size=36,
+        frame_size=(400, 100)))
+
+    return results
+
+
+def test_multiline_fancy():
+    """Test multiline with various styles."""
+    print("\n--- Multiline Fancy ---")
+    results = []
+
+    # Right-aligned (1px tolerance: same sub-pixel issue as center alignment)
+    results.append(compare("multi_right",
+        text="Right\nAligned\nText", x=380, y=20,
+        frame_size=(400, 150),
+        multiline=True, anchor="ra", align="right",
+        tolerance=1))
+
+    # Center + stroke
+    results.append(compare("multi_center_stroke",
+        text="Title\nSubtitle", x=200, y=20,
+        font_size=32, frame_size=(400, 120),
+        multiline=True, anchor="ma", align="center",
+        stroke_width=2, stroke_fill=(0, 0, 0),
+        tolerance=1))
+
+    # Wide line spacing
+    results.append(compare("multi_wide_spacing",
+        text="Line A\nLine B\nLine C", x=20, y=10,
+        frame_size=(300, 200),
+        multiline=True, line_spacing=20))
+
+    # Zero extra spacing
+    results.append(compare("multi_tight_spacing",
+        text="Tight\nPacked\nLines", x=20, y=10,
+        frame_size=(300, 150),
+        multiline=True, line_spacing=0))
+
+    # Many lines
+    results.append(compare("multi_many_lines",
+        text="One\nTwo\nThree\nFour\nFive\nSix", x=20, y=5,
+        font_size=20, frame_size=(200, 200),
+        multiline=True, line_spacing=4))
+
+    # Bold multiline with stroke
+    results.append(compare("multi_bold_stroke",
+        text="BOLD\nSTROKE", x=20, y=10,
+        font_path=FONTS['bold'], font_size=48,
+        stroke_width=3, stroke_fill=(200, 0, 0),
+        frame_size=(350, 150), multiline=True))
+
+    # Multiline on colored bg with opacity
+    results.append(compare("multi_opacity_on_bg",
+        text="Semi\nTransparent", x=20, y=10,
+        fill=(255, 255, 0), bg=(0, 50, 100), opacity=0.7,
+        frame_size=(300, 120), multiline=True))
+
+    return results
+
+
+def test_special_chars():
+    """Test special characters and edge cases."""
+    print("\n--- Special Characters ---")
+    results = []
+
+    # Numbers and symbols
+    results.append(compare("chars_numbers",
+        text="0123456789", x=20, y=30, frame_size=(300, 80)))
+
+    results.append(compare("chars_punctuation",
+        text="Hello, World! (v2.0)", x=10, y=30, frame_size=(350, 80)))
+
+    results.append(compare("chars_symbols",
+        text="@#$%^&*+=", x=20, y=30, frame_size=(300, 80)))
+
+    # Single character
+    results.append(compare("chars_single",
+        text="X", x=50, y=30, font_size=48, frame_size=(100, 80)))
+
+    # Very long text (clipped)
+    results.append(compare("chars_long",
+        text="The quick brown fox jumps over the lazy dog", x=10, y=30,
+        font_size=24, frame_size=(400, 80)))
+
+    # Mixed case
+    results.append(compare("chars_mixed_case",
+        text="AaBbCcDdEeFf", x=10, y=30, frame_size=(350, 80)))
+
+    return results
+
+
+def test_combos():
+    """Complex combinations of features."""
+    print("\n--- Combos ---")
+    results = []
+
+    # Big bold stroke + color + opacity
+    results.append(compare("combo_all_features",
+        text="EPIC", x=20, y=10,
+        font_path=FONTS['bold'], font_size=64,
+        fill=(255, 200, 0), bg=(20, 0, 40), opacity=0.85,
+        stroke_width=4, stroke_fill=(180, 0, 0),
+        frame_size=(350, 100)))
+
+    # Small mono on busy background
+    results.append(compare("combo_mono_code",
+        text="fn main() {}", x=10, y=15,
+        font_path=FONTS['mono'], font_size=16,
+        fill=(0, 255, 100), bg=(30, 30, 30),
+        frame_size=(250, 50)))
+
+    # Serif italic multiline with stroke
+    results.append(compare("combo_serif_italic_multi",
+        text="Once upon\na time...", x=20, y=10,
+        font_path=FONTS['italic'], font_size=28,
+        stroke_width=1, stroke_fill=(80, 80, 80),
+        frame_size=(300, 120), multiline=True))
+
+    # Narrow font, big stroke, center anchored
+    results.append(compare("combo_narrow_stroke_center",
+        text="NARROW", x=150, y=40,
+        font_path=FONTS['narrow'], font_size=44,
+        stroke_width=5, stroke_fill=(0, 0, 0),
+        anchor="mm", frame_size=(300, 80)))
+
+    # Multiple strips on same frame (simulated by sequential placement)
+    results.append(compare("combo_opacity_blend",
+        text="Layered", x=20, y=30,
+        fill=(255, 0, 0), bg=(0, 0, 255), opacity=0.5,
+        font_size=48, frame_size=(300, 80)))
+
+    return results
+
+
+def test_multi_strip_overlay():
+    """Test placing multiple strips on the same frame."""
+    print("\n--- Multi-Strip Overlay ---")
+    results = []
+
+    frame_size = (400, 150)
+    bg = (20, 20, 40)
+
+    # PIL version - multiple draw calls
+    font1 = _load_font(FONTS['bold'], 48)
+    font2 = _load_font(None, 24)
+    font3 = _load_font(FONTS['mono'], 18)
+
+    pil_frame = np.full((frame_size[1], frame_size[0], 3), bg, dtype=np.uint8)
+    txt_layer = Image.new('RGBA', frame_size, (0, 0, 0, 0))
+    draw = ImageDraw.Draw(txt_layer)
+    draw.text((20, 10), "TITLE", fill=(255, 255, 0, 255), font=font1,
+              stroke_width=2, stroke_fill=(0, 0, 0, 255))
+    draw.text((20, 70), "Subtitle here", fill=(200, 200, 200, 255), font=font2)
+    draw.text((20, 110), "code_snippet()", fill=(0, 255, 128, 200), font=font3)
+    bg_img = Image.fromarray(pil_frame).convert('RGBA')
+    pil_result = np.array(Image.alpha_composite(bg_img, txt_layer).convert('RGB'))
+
+    # Strip version - multiple placements
+    frame = jnp.full((frame_size[1], frame_size[0], 3), jnp.array(bg, dtype=jnp.uint8), dtype=jnp.uint8)
+
+    s1 = render_text_strip("TITLE", FONTS['bold'], 48, stroke_width=2, stroke_fill=(0, 0, 0))
+    s2 = render_text_strip("Subtitle here", None, 24)
+    s3 = render_text_strip("code_snippet()", FONTS['mono'], 18)
+
+    frame = place_text_strip_jax(
+        frame, jnp.asarray(s1.image), 20, 10,
+        s1.baseline_y, s1.bearing_x,
+        jnp.array([255, 255, 0], dtype=jnp.float32), 1.0,
+        anchor_x=s1.anchor_x, anchor_y=s1.anchor_y,
+        stroke_width=s1.stroke_width)
+
+    frame = place_text_strip_jax(
+        frame, jnp.asarray(s2.image), 20, 70,
+        s2.baseline_y, s2.bearing_x,
+        jnp.array([200, 200, 200], dtype=jnp.float32), 1.0,
+        anchor_x=s2.anchor_x, anchor_y=s2.anchor_y,
+        stroke_width=s2.stroke_width)
+
+    frame = place_text_strip_jax(
+        frame, jnp.asarray(s3.image), 20, 110,
+        s3.baseline_y, s3.bearing_x,
+        jnp.array([0, 255, 128], dtype=jnp.float32), 200/255,
+        anchor_x=s3.anchor_x, anchor_y=s3.anchor_y,
+        stroke_width=s3.stroke_width)
+
+    strip_result = np.array(frame)
+
+    diff = np.abs(pil_result.astype(np.int16) - strip_result.astype(np.int16))
+    max_diff = diff.max()
+    pixels_diff = (diff > 0).any(axis=2).sum()
+
+    if max_diff <= 1:
+        print(f"  PASS: multi_overlay (max_diff={max_diff})")
+        results.append(True)
+    else:
+        print(f"  FAIL: multi_overlay")
+        print(f"        Max diff: {max_diff}, Pixels different: {pixels_diff}")
+        Image.fromarray(pil_result).save("/tmp/pil_multi_overlay.png")
+        Image.fromarray(strip_result).save("/tmp/strip_multi_overlay.png")
+        diff_vis = np.clip(diff * 10, 0, 255).astype(np.uint8)
+        Image.fromarray(diff_vis).save("/tmp/diff_multi_overlay.png")
+        print(f"        Saved: /tmp/pil_multi_overlay.png /tmp/strip_multi_overlay.png /tmp/diff_multi_overlay.png")
+        results.append(False)
+
+    return results
+
+
+def main():
+    print("=" * 60)
+    print("Funky TextStrip vs PIL Comparison")
+    print("=" * 60)
+
+    all_results = []
+    all_results.extend(test_colors())
+    all_results.extend(test_opacity())
+    all_results.extend(test_fonts())
+    all_results.extend(test_anchors())
+    all_results.extend(test_strokes())
+    all_results.extend(test_edge_clipping())
+    all_results.extend(test_multiline_fancy())
+    all_results.extend(test_special_chars())
+    all_results.extend(test_combos())
+    all_results.extend(test_multi_strip_overlay())
+
+    print("\n" + "=" * 60)
+    passed = sum(all_results)
+    total = len(all_results)
+    print(f"Results: {passed}/{total} passed")
+    if passed == total:
+        print("ALL TESTS PASSED!")
+    else:
+        failed = [i for i, r in enumerate(all_results) if not r]
+        print(f"FAILED: {total - passed} tests (indices: {failed})")
+    print("=" * 60)
+
+
+if __name__ == "__main__":
+    main()

test_pil_options.py

@@ -0,0 +1,183 @@
+#!/usr/bin/env python3
+"""
+Explore PIL text options and test if we can match them.
+"""
+
+import numpy as np
+from PIL import Image, ImageDraw, ImageFont
+
+def load_font(font_name=None, font_size=32):
+    """Load a font."""
+    candidates = [
+        font_name,
+        '/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf',
+        '/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf',
+        '/usr/share/fonts/truetype/dejavu/DejaVuSans-Oblique.ttf',
+    ]
+    for path in candidates:
+        if path is None:
+            continue
+        try:
+            return ImageFont.truetype(path, font_size)
+        except (IOError, OSError):
+            continue
+    return ImageFont.load_default()
+
+
+def test_pil_options():
+    """Test various PIL text options."""
+
+    # Create a test frame
+    frame_size = (600, 400)
+
+    font = load_font(None, 36)
+    font_bold = load_font('/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf', 36)
+    font_italic = load_font('/usr/share/fonts/truetype/dejavu/DejaVuSans-Oblique.ttf', 36)
+
+    tests = []
+
+    # Test 1: Basic text
+    def basic_text():
+        img = Image.new('RGBA', frame_size, (0, 0, 0, 0))
+        draw = ImageDraw.Draw(img)
+        draw.text((20, 20), "Basic Text", fill=(255, 255, 255, 255), font=font)
+        return img, "basic"
+    tests.append(basic_text)
+
+    # Test 2: Stroke/outline
+    def stroke_text():
+        img = Image.new('RGBA', frame_size, (0, 0, 0, 0))
+        draw = ImageDraw.Draw(img)
+        draw.text((20, 20), "Stroke Text", fill=(255, 255, 255, 255), font=font,
+                  stroke_width=2, stroke_fill=(255, 0, 0, 255))
+        return img, "stroke"
+    tests.append(stroke_text)
+
+    # Test 3: Bold font
+    def bold_text():
+        img = Image.new('RGBA', frame_size, (0, 0, 0, 0))
+        draw = ImageDraw.Draw(img)
+        draw.text((20, 20), "Bold Text", fill=(255, 255, 255, 255), font=font_bold)
+        return img, "bold"
+    tests.append(bold_text)
+
+    # Test 4: Italic font
+    def italic_text():
+        img = Image.new('RGBA', frame_size, (0, 0, 0, 0))
+        draw = ImageDraw.Draw(img)
+        draw.text((20, 20), "Italic Text", fill=(255, 255, 255, 255), font=font_italic)
+        return img, "italic"
+    tests.append(italic_text)
+
+    # Test 5: Different anchors
+    def anchor_text():
+        img = Image.new('RGBA', frame_size, (0, 0, 0, 0))
+        draw = ImageDraw.Draw(img)
+        # Draw crosshairs at anchor points
+        for x in [100, 300, 500]:
+            draw.line([(x-10, 50), (x+10, 50)], fill=(100, 100, 100, 255))
+            draw.line([(x, 40), (x, 60)], fill=(100, 100, 100, 255))
+
+        draw.text((100, 50), "Left", fill=(255, 255, 255, 255), font=font, anchor="lm")
+        draw.text((300, 50), "Center", fill=(255, 255, 255, 255), font=font, anchor="mm")
+        draw.text((500, 50), "Right", fill=(255, 255, 255, 255), font=font, anchor="rm")
+        return img, "anchor"
+    tests.append(anchor_text)
+
+    # Test 6: Multiline text
+    def multiline_text():
+        img = Image.new('RGBA', frame_size, (0, 0, 0, 0))
+        draw = ImageDraw.Draw(img)
+        draw.multiline_text((20, 20), "Line One\nLine Two\nLine Three",
+                            fill=(255, 255, 255, 255), font=font, spacing=10)
+        return img, "multiline"
+    tests.append(multiline_text)
+
+    # Test 7: Semi-transparent text
+    def alpha_text():
+        img = Image.new('RGBA', frame_size, (0, 0, 0, 0))
+        draw = ImageDraw.Draw(img)
+        draw.text((20, 20), "Alpha 100%", fill=(255, 255, 255, 255), font=font)
+        draw.text((20, 60), "Alpha 50%", fill=(255, 255, 255, 128), font=font)
+        draw.text((20, 100), "Alpha 25%", fill=(255, 255, 255, 64), font=font)
+        return img, "alpha"
+    tests.append(alpha_text)
+
+    # Test 8: Colored text
+    def colored_text():
+        img = Image.new('RGBA', frame_size, (0, 0, 0, 0))
+        draw = ImageDraw.Draw(img)
+        draw.text((20, 20), "Red", fill=(255, 0, 0, 255), font=font)
+        draw.text((20, 60), "Green", fill=(0, 255, 0, 255), font=font)
+        draw.text((20, 100), "Blue", fill=(0, 0, 255, 255), font=font)
+        draw.text((20, 140), "Yellow", fill=(255, 255, 0, 255), font=font)
+        return img, "colored"
+    tests.append(colored_text)
+
+    # Test 9: Large stroke
+    def large_stroke():
+        img = Image.new('RGBA', frame_size, (0, 0, 0, 0))
+        draw = ImageDraw.Draw(img)
+        draw.text((20, 20), "Big Stroke", fill=(255, 255, 255, 255), font=font,
+                  stroke_width=5, stroke_fill=(0, 0, 0, 255))
+        return img, "large_stroke"
+    tests.append(large_stroke)
+
+    # Test 10: Emoji (if supported)
+    def emoji_text():
+        img = Image.new('RGBA', frame_size, (0, 0, 0, 0))
+        draw = ImageDraw.Draw(img)
+        try:
+            # Try to find an emoji font
+            emoji_font = None
+            emoji_paths = [
+                '/usr/share/fonts/truetype/noto/NotoColorEmoji.ttf',
+                '/usr/share/fonts/truetype/ancient-scripts/Symbola_hint.ttf',
+            ]
+            for p in emoji_paths:
+                try:
+                    emoji_font = ImageFont.truetype(p, 36)
+                    break
+                except:
+                    pass
+            if emoji_font:
+                draw.text((20, 20), "Hello 🎵 World 🎸", fill=(255, 255, 255, 255), font=emoji_font)
+            else:
+                draw.text((20, 20), "No emoji font found", fill=(255, 255, 255, 255), font=font)
+        except Exception as e:
+            draw.text((20, 20), f"Emoji error: {e}", fill=(255, 255, 255, 255), font=font)
+        return img, "emoji"
+    tests.append(emoji_text)
+
+    # Run all tests
+    print("PIL Text Options Test")
+    print("=" * 60)
+
+    for test_fn in tests:
+        img, name = test_fn()
+        fname = f"/tmp/pil_test_{name}.png"
+        img.save(fname)
+        print(f"Saved: {fname}")
+
+    print("\nCheck /tmp/pil_test_*.png for results")
+
+    # Print available parameters
+    print("\n" + "=" * 60)
+    print("PIL draw.text() parameters:")
+    print("  - xy: position tuple")
+    print("  - text: string to draw")
+    print("  - fill: color (R,G,B) or (R,G,B,A)")
+    print("  - font: ImageFont object")
+    print("  - anchor: 2-char code (la=left-ascender, mm=middle-middle, etc.)")
+    print("  - spacing: line spacing for multiline")
+    print("  - align: 'left', 'center', 'right' for multiline")
+    print("  - direction: 'rtl', 'ltr', 'ttb' (requires libraqm)")
+    print("  - features: OpenType features list")
+    print("  - language: language code for shaping")
+    print("  - stroke_width: outline width in pixels")
+    print("  - stroke_fill: outline color")
+    print("  - embedded_color: use embedded color glyphs (emoji)")
+
+
+if __name__ == "__main__":
+    test_pil_options()

test_styled_text.py

@@ -0,0 +1,176 @@
+#!/usr/bin/env python3
+"""
+Test styled TextStrip rendering against PIL.
+"""
+
+import numpy as np
+import jax.numpy as jnp
+from PIL import Image, ImageDraw, ImageFont
+
+from streaming.jax_typography import (
+    render_text_strip, place_text_strip_jax, _load_font
+)
+
+
+def render_pil(text, x, y, font_size=36, frame_size=(400, 100),
+               stroke_width=0, stroke_fill=None, anchor="la",
+               multiline=False, line_spacing=4, align="left"):
+    """Render with PIL directly."""
+    frame = np.zeros((frame_size[1], frame_size[0], 3), dtype=np.uint8)
+    img = Image.fromarray(frame)
+    draw = ImageDraw.Draw(img)
+
+    font = _load_font(None, font_size)
+
+    # Default stroke fill
+    if stroke_fill is None:
+        stroke_fill = (0, 0, 0)
+
+    if multiline:
+        draw.multiline_text((x, y), text, fill=(255, 255, 255), font=font,
+                           stroke_width=stroke_width, stroke_fill=stroke_fill,
+                           spacing=line_spacing, align=align, anchor=anchor)
+    else:
+        draw.text((x, y), text, fill=(255, 255, 255), font=font,
+                  stroke_width=stroke_width, stroke_fill=stroke_fill, anchor=anchor)
+
+    return np.array(img)
+
+
+def render_strip(text, x, y, font_size=36, frame_size=(400, 100),
+                 stroke_width=0, stroke_fill=None, anchor="la",
+                 multiline=False, line_spacing=4, align="left"):
+    """Render with TextStrip."""
+    frame = jnp.zeros((frame_size[1], frame_size[0], 3), dtype=jnp.uint8)
+
+    strip = render_text_strip(
+        text, None, font_size,
+        stroke_width=stroke_width, stroke_fill=stroke_fill,
+        anchor=anchor, multiline=multiline, line_spacing=line_spacing, align=align
+    )
+    strip_img = jnp.asarray(strip.image)
+    color = jnp.array([255, 255, 255], dtype=jnp.float32)
+
+    result = place_text_strip_jax(
+        frame, strip_img, x, y,
+        strip.baseline_y, strip.bearing_x,
+        color, 1.0,
+        anchor_x=strip.anchor_x, anchor_y=strip.anchor_y,
+        stroke_width=strip.stroke_width
+    )
+
+    return np.array(result)
+
+
+def compare(name, text, x, y, font_size=36, frame_size=(400, 100),
+            tolerance=0, **kwargs):
+    """Compare PIL and TextStrip rendering.
+
+    tolerance=0: exact pixel match required
+    tolerance=1: allow 1-pixel position shift (for sub-pixel rendering differences
+                 in center-aligned multiline text where the strip is pre-rendered
+                 at a different base position than the final placement)
+    """
+    pil = render_pil(text, x, y, font_size, frame_size, **kwargs)
+    strip = render_strip(text, x, y, font_size, frame_size, **kwargs)
+
+    diff = np.abs(pil.astype(np.int16) - strip.astype(np.int16))
+    max_diff = diff.max()
+    pixels_diff = (diff > 0).any(axis=2).sum()
+
+    if max_diff == 0:
+        print(f"PASS: {name}")
+        print(f"       Max diff: 0, Pixels different: 0")
+        return True
+
+    if tolerance > 0:
+        # Check if the difference is just a sub-pixel position shift:
+        # for each shifted version, compute the minimum diff
+        best_diff = diff.copy()
+        for dy in range(-tolerance, tolerance + 1):
+            for dx in range(-tolerance, tolerance + 1):
+                if dy == 0 and dx == 0:
+                    continue
+                shifted = np.roll(np.roll(strip, dy, axis=0), dx, axis=1)
+                sdiff = np.abs(pil.astype(np.int16) - shifted.astype(np.int16))
+                best_diff = np.minimum(best_diff, sdiff)
+        max_shift_diff = best_diff.max()
+        pixels_shift_diff = (best_diff > 0).any(axis=2).sum()
+        if max_shift_diff == 0:
+            print(f"PASS: {name} (within {tolerance}px position tolerance)")
+            print(f"       Raw diff: {max_diff}, After shift tolerance: 0")
+            return True
+
+    status = "FAIL"
+    print(f"{status}: {name}")
+    print(f"       Max diff: {max_diff}, Pixels different: {pixels_diff}")
+
+    # Save debug images
+    Image.fromarray(pil).save(f"/tmp/pil_{name}.png")
+    Image.fromarray(strip).save(f"/tmp/strip_{name}.png")
+    diff_scaled = np.clip(diff * 10, 0, 255).astype(np.uint8)
+    Image.fromarray(diff_scaled).save(f"/tmp/diff_{name}.png")
+    print(f"       Saved: /tmp/pil_{name}.png, /tmp/strip_{name}.png, /tmp/diff_{name}.png")
+
+    return False
+
+
+def main():
+    print("=" * 60)
+    print("Styled TextStrip vs PIL Comparison")
+    print("=" * 60)
+
+    results = []
+
+    # Basic text
+    results.append(compare("basic", "Hello World", 20, 50))
+
+    # Stroke/outline
+    results.append(compare("stroke_2", "Outlined", 20, 50,
+                          stroke_width=2, stroke_fill=(255, 0, 0)))
+
+    results.append(compare("stroke_5", "Big Outline", 30, 60, font_size=48,
+                          frame_size=(500, 120),
+                          stroke_width=5, stroke_fill=(0, 0, 0)))
+
+    # Anchors - center
+    results.append(compare("anchor_mm", "Center", 200, 50, frame_size=(400, 100),
+                          anchor="mm"))
+
+    # Anchors - right
+    results.append(compare("anchor_rm", "Right", 380, 50, frame_size=(400, 100),
+                          anchor="rm"))
+
+    # Multiline
+    results.append(compare("multiline", "Line 1\nLine 2\nLine 3", 20, 20,
+                          frame_size=(400, 150),
+                          multiline=True, line_spacing=8))
+
+    # Multiline centered (1px tolerance: sub-pixel rendering differs because
+    # the strip is pre-rendered at an integer position while PIL's center
+    # alignment uses fractional getlength values for the 'm' anchor shift)
+    results.append(compare("multiline_center", "Short\nMedium Length\nX", 200, 20,
+                          frame_size=(400, 150),
+                          multiline=True, anchor="ma", align="center",
+                          tolerance=1))
+
+    # Stroke + multiline
+    results.append(compare("stroke_multiline", "Line A\nLine B", 20, 20,
+                          frame_size=(400, 120),
+                          stroke_width=2, stroke_fill=(0, 0, 255),
+                          multiline=True))
+
+    print("=" * 60)
+    passed = sum(results)
+    total = len(results)
+    print(f"Results: {passed}/{total} passed")
+
+    if passed == total:
+        print("ALL TESTS PASSED!")
+    else:
+        print(f"FAILED: {total - passed} tests")
+    print("=" * 60)
+
+
+if __name__ == "__main__":
+    main()

tests/test_jax_pipeline_integration.py

@@ -0,0 +1,517 @@
+#!/usr/bin/env python3
+"""Integration tests comparing JAX and Python rendering pipelines.
+
+These tests ensure the JAX-compiled effect chains produce identical output
+to the Python/NumPy path. They test:
+1. Full effect pipelines through both interpreters
+2. Multi-frame sequences (to catch phase-dependent bugs)
+3. Compiled effect chain fusion
+4. Edge cases like shrinking/zooming that affect boundary handling
+"""
+import os
+import sys
+import pytest
+import numpy as np
+import shutil
+from pathlib import Path
+
+# Ensure the art-celery module is importable
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+from sexp_effects.primitive_libs import core as core_mod
+
+
+# Path to test resources
+TEST_DIR = Path('/home/giles/art/test')
+EFFECTS_DIR = TEST_DIR / 'sexp_effects' / 'effects'
+TEMPLATES_DIR = TEST_DIR / 'templates'
+
+
+def create_test_image(h=96, w=128):
+    """Create a test image with distinct patterns."""
+    import cv2
+    img = np.zeros((h, w, 3), dtype=np.uint8)
+
+    # Create gradient background
+    for y in range(h):
+        for x in range(w):
+            img[y, x] = [
+                int(255 * x / w),  # R: horizontal gradient
+                int(255 * y / h),  # G: vertical gradient
+                128  # B: constant
+            ]
+
+    # Add features
+    cv2.circle(img, (w//2, h//2), 20, (255, 0, 0), -1)
+    cv2.rectangle(img, (10, 10), (30, 30), (0, 255, 0), -1)
+
+    return img
+
+
+@pytest.fixture(scope='module')
+def test_env(tmp_path_factory):
+    """Set up test environment with sexp files and test media."""
+    test_dir = tmp_path_factory.mktemp('sexp_test')
+    original_dir = os.getcwd()
+    os.chdir(test_dir)
+
+    # Create directories
+    (test_dir / 'effects').mkdir()
+    (test_dir / 'sexp_effects' / 'effects').mkdir(parents=True)
+
+    # Create test image
+    import cv2
+    test_img = create_test_image()
+    cv2.imwrite(str(test_dir / 'test_image.png'), test_img)
+
+    # Copy required effect files
+    for effect in ['rotate', 'zoom', 'blend', 'invert', 'hue_shift']:
+        src = EFFECTS_DIR / f'{effect}.sexp'
+        dst = test_dir / 'sexp_effects' / 'effects' / f'{effect}.sexp'
+        if src.exists():
+            shutil.copy(src, dst)
+
+    yield {
+        'dir': test_dir,
+        'image_path': test_dir / 'test_image.png',
+        'test_img': test_img,
+    }
+
+    os.chdir(original_dir)
+
+
+def create_sexp_file(test_dir, content, filename='test.sexp'):
+    """Create a test sexp file."""
+    path = test_dir / 'effects' / filename
+    with open(path, 'w') as f:
+        f.write(content)
+    return str(path)
+
+
+class TestJaxPythonPipelineEquivalence:
+    """Test that JAX and Python pipelines produce equivalent output."""
+
+    def test_single_rotate_effect(self, test_env):
+        """Test that a single rotate effect matches between Python and JAX."""
+        sexp_content = '''(stream "test"
+  :width 128
+  :height 96
+  :seed 42
+
+  (effect rotate :path "../sexp_effects/effects/rotate.sexp")
+
+  (frame (rotate frame :angle 15 :speed 0))
+)
+'''
+        sexp_path = create_sexp_file(test_env['dir'], sexp_content)
+
+        from streaming.stream_sexp_generic import StreamInterpreter, Context
+        import cv2
+
+        test_img = cv2.imread(str(test_env['image_path']))
+
+        # Python path
+        core_mod.set_random_seed(42)
+        py_interp = StreamInterpreter(sexp_path, use_jax=False)
+        py_interp._init()
+
+        # JAX path
+        core_mod.set_random_seed(42)
+        jax_interp = StreamInterpreter(sexp_path, use_jax=True)
+        jax_interp._init()
+
+        ctx = Context(fps=10)
+        ctx.t = 0.5
+        ctx.frame_num = 5
+
+        frame_env = {
+            'ctx': {'t': ctx.t, 'frame-num': ctx.frame_num, 'fps': 10},
+            't': ctx.t, 'frame-num': ctx.frame_num,
+        }
+
+        # Inject test image into globals
+        py_interp.globals['frame'] = test_img
+        jax_interp.globals['frame'] = test_img
+
+        py_result = py_interp._eval(py_interp.frame_pipeline, frame_env)
+        jax_result = jax_interp._eval(jax_interp.frame_pipeline, frame_env)
+
+        # Force deferred if needed
+        py_result = np.asarray(py_interp._maybe_force(py_result))
+        jax_result = np.asarray(jax_interp._maybe_force(jax_result))
+
+        diff = np.abs(py_result.astype(float) - jax_result.astype(float))
+        mean_diff = np.mean(diff)
+
+        assert mean_diff < 2.0, f"Rotate effect: mean diff {mean_diff:.2f} exceeds threshold"
+
+    def test_rotate_then_zoom(self, test_env):
+        """Test rotate followed by zoom - tests effect chain fusion."""
+        sexp_content = '''(stream "test"
+  :width 128
+  :height 96
+  :seed 42
+
+  (effect rotate :path "../sexp_effects/effects/rotate.sexp")
+  (effect zoom :path "../sexp_effects/effects/zoom.sexp")
+
+  (frame (-> (rotate frame :angle 15 :speed 0)
+             (zoom :amount 0.95 :speed 0)))
+)
+'''
+        sexp_path = create_sexp_file(test_env['dir'], sexp_content)
+
+        from streaming.stream_sexp_generic import StreamInterpreter, Context
+        import cv2
+
+        test_img = cv2.imread(str(test_env['image_path']))
+
+        core_mod.set_random_seed(42)
+        py_interp = StreamInterpreter(sexp_path, use_jax=False)
+        py_interp._init()
+
+        core_mod.set_random_seed(42)
+        jax_interp = StreamInterpreter(sexp_path, use_jax=True)
+        jax_interp._init()
+
+        ctx = Context(fps=10)
+        ctx.t = 0.5
+        ctx.frame_num = 5
+
+        frame_env = {
+            'ctx': {'t': ctx.t, 'frame-num': ctx.frame_num, 'fps': 10},
+            't': ctx.t, 'frame-num': ctx.frame_num,
+        }
+
+        py_interp.globals['frame'] = test_img
+        jax_interp.globals['frame'] = test_img
+
+        py_result = np.asarray(py_interp._maybe_force(
+            py_interp._eval(py_interp.frame_pipeline, frame_env)))
+        jax_result = np.asarray(jax_interp._maybe_force(
+            jax_interp._eval(jax_interp.frame_pipeline, frame_env)))
+
+        diff = np.abs(py_result.astype(float) - jax_result.astype(float))
+        mean_diff = np.mean(diff)
+
+        assert mean_diff < 2.0, f"Rotate+zoom chain: mean diff {mean_diff:.2f} exceeds threshold"
+
+    def test_zoom_shrink_boundary_handling(self, test_env):
+        """Test zoom with shrinking factor - critical for boundary handling."""
+        sexp_content = '''(stream "test"
+  :width 128
+  :height 96
+  :seed 42
+
+  (effect zoom :path "../sexp_effects/effects/zoom.sexp")
+
+  (frame (zoom frame :amount 0.8 :speed 0))
+)
+'''
+        sexp_path = create_sexp_file(test_env['dir'], sexp_content)
+
+        from streaming.stream_sexp_generic import StreamInterpreter, Context
+        import cv2
+
+        test_img = cv2.imread(str(test_env['image_path']))
+
+        core_mod.set_random_seed(42)
+        py_interp = StreamInterpreter(sexp_path, use_jax=False)
+        py_interp._init()
+
+        core_mod.set_random_seed(42)
+        jax_interp = StreamInterpreter(sexp_path, use_jax=True)
+        jax_interp._init()
+
+        ctx = Context(fps=10)
+        ctx.t = 0.5
+        ctx.frame_num = 5
+
+        frame_env = {
+            'ctx': {'t': ctx.t, 'frame-num': ctx.frame_num, 'fps': 10},
+            't': ctx.t, 'frame-num': ctx.frame_num,
+        }
+
+        py_interp.globals['frame'] = test_img
+        jax_interp.globals['frame'] = test_img
+
+        py_result = np.asarray(py_interp._maybe_force(
+            py_interp._eval(py_interp.frame_pipeline, frame_env)))
+        jax_result = np.asarray(jax_interp._maybe_force(
+            jax_interp._eval(jax_interp.frame_pipeline, frame_env)))
+
+        # Check corners specifically - these are most affected by boundary handling
+        h, w = test_img.shape[:2]
+        corners = [(0, 0), (0, w-1), (h-1, 0), (h-1, w-1)]
+        for y, x in corners:
+            py_val = py_result[y, x]
+            jax_val = jax_result[y, x]
+            corner_diff = np.abs(py_val.astype(float) - jax_val.astype(float)).max()
+            assert corner_diff < 10.0, f"Corner ({y},{x}): diff {corner_diff} - py={py_val}, jax={jax_val}"
+
+    def test_blend_two_clips(self, test_env):
+        """Test blending two effect chains - the core bug scenario."""
+        sexp_content = '''(stream "test"
+  :width 128
+  :height 96
+  :seed 42
+
+  (require-primitives "core")
+  (require-primitives "image")
+  (require-primitives "blending")
+  (effect rotate :path "../sexp_effects/effects/rotate.sexp")
+  (effect zoom :path "../sexp_effects/effects/zoom.sexp")
+  (effect blend :path "../sexp_effects/effects/blend.sexp")
+
+  (frame
+    (let [clip_a (-> (rotate frame :angle 5 :speed 0)
+                     (zoom :amount 1.05 :speed 0))
+          clip_b (-> (rotate frame :angle -5 :speed 0)
+                     (zoom :amount 0.95 :speed 0))]
+      (blend :base clip_a :overlay clip_b :opacity 0.5)))
+)
+'''
+        sexp_path = create_sexp_file(test_env['dir'], sexp_content)
+
+        from streaming.stream_sexp_generic import StreamInterpreter, Context
+        import cv2
+
+        test_img = cv2.imread(str(test_env['image_path']))
+
+        core_mod.set_random_seed(42)
+        py_interp = StreamInterpreter(sexp_path, use_jax=False)
+        py_interp._init()
+
+        core_mod.set_random_seed(42)
+        jax_interp = StreamInterpreter(sexp_path, use_jax=True)
+        jax_interp._init()
+
+        ctx = Context(fps=10)
+        ctx.t = 0.5
+        ctx.frame_num = 5
+
+        frame_env = {
+            'ctx': {'t': ctx.t, 'frame-num': ctx.frame_num, 'fps': 10},
+            't': ctx.t, 'frame-num': ctx.frame_num,
+        }
+
+        py_interp.globals['frame'] = test_img
+        jax_interp.globals['frame'] = test_img
+
+        py_result = np.asarray(py_interp._maybe_force(
+            py_interp._eval(py_interp.frame_pipeline, frame_env)))
+        jax_result = np.asarray(jax_interp._maybe_force(
+            jax_interp._eval(jax_interp.frame_pipeline, frame_env)))
+
+        diff = np.abs(py_result.astype(float) - jax_result.astype(float))
+        mean_diff = np.mean(diff)
+        max_diff = np.max(diff)
+
+        # Check edge region specifically
+        edge_diff = diff[0, :].mean()
+
+        assert mean_diff < 3.0, f"Blend: mean diff {mean_diff:.2f} exceeds threshold"
+        assert edge_diff < 10.0, f"Blend edge: diff {edge_diff:.2f} exceeds threshold"
+
+    def test_blend_with_invert(self, test_env):
+        """Test blending with invert - matches the problematic recipe pattern."""
+        sexp_content = '''(stream "test"
+  :width 128
+  :height 96
+  :seed 42
+
+  (require-primitives "core")
+  (require-primitives "image")
+  (require-primitives "blending")
+  (require-primitives "color_ops")
+  (effect rotate :path "../sexp_effects/effects/rotate.sexp")
+  (effect zoom :path "../sexp_effects/effects/zoom.sexp")
+  (effect blend :path "../sexp_effects/effects/blend.sexp")
+  (effect invert :path "../sexp_effects/effects/invert.sexp")
+
+  (frame
+    (let [clip_a (-> (rotate frame :angle 5 :speed 0)
+                     (zoom :amount 1.05 :speed 0)
+                     (invert :amount 1))
+          clip_b (-> (rotate frame :angle -5 :speed 0)
+                     (zoom :amount 0.95 :speed 0))]
+      (blend :base clip_a :overlay clip_b :opacity 0.5)))
+)
+'''
+        sexp_path = create_sexp_file(test_env['dir'], sexp_content)
+
+        from streaming.stream_sexp_generic import StreamInterpreter, Context
+        import cv2
+
+        test_img = cv2.imread(str(test_env['image_path']))
+
+        core_mod.set_random_seed(42)
+        py_interp = StreamInterpreter(sexp_path, use_jax=False)
+        py_interp._init()
+
+        core_mod.set_random_seed(42)
+        jax_interp = StreamInterpreter(sexp_path, use_jax=True)
+        jax_interp._init()
+
+        ctx = Context(fps=10)
+        ctx.t = 0.5
+        ctx.frame_num = 5
+
+        frame_env = {
+            'ctx': {'t': ctx.t, 'frame-num': ctx.frame_num, 'fps': 10},
+            't': ctx.t, 'frame-num': ctx.frame_num,
+        }
+
+        py_interp.globals['frame'] = test_img
+        jax_interp.globals['frame'] = test_img
+
+        py_result = np.asarray(py_interp._maybe_force(
+            py_interp._eval(py_interp.frame_pipeline, frame_env)))
+        jax_result = np.asarray(jax_interp._maybe_force(
+            jax_interp._eval(jax_interp.frame_pipeline, frame_env)))
+
+        diff = np.abs(py_result.astype(float) - jax_result.astype(float))
+        mean_diff = np.mean(diff)
+
+        assert mean_diff < 3.0, f"Blend+invert: mean diff {mean_diff:.2f} exceeds threshold"
+
+
+class TestDeferredEffectChainFusion:
+    """Test the DeferredEffectChain fusion mechanism specifically."""
+
+    def test_manual_vs_fused_chain(self, test_env):
+        """Compare manual application vs fused DeferredEffectChain."""
+        import jax.numpy as jnp
+        from streaming.stream_sexp_generic import StreamInterpreter, DeferredEffectChain
+
+        # Create minimal sexp to load effects
+        sexp_content = '''(stream "test"
+  :width 128
+  :height 96
+  :seed 42
+
+  (effect rotate :path "../sexp_effects/effects/rotate.sexp")
+  (effect zoom :path "../sexp_effects/effects/zoom.sexp")
+
+  (frame frame)
+)
+'''
+        sexp_path = create_sexp_file(test_env['dir'], sexp_content)
+
+        core_mod.set_random_seed(42)
+        interp = StreamInterpreter(sexp_path, use_jax=True)
+        interp._init()
+
+        test_img = test_env['test_img']
+        jax_frame = jnp.array(test_img)
+        t = 0.5
+        frame_num = 5
+
+        # Manual step-by-step application
+        rotate_fn = interp.jax_effects['rotate']
+        zoom_fn = interp.jax_effects['zoom']
+
+        rot_angle = -5.0
+        zoom_amount = 0.95
+
+        manual_result = rotate_fn(jax_frame, t=t, frame_num=frame_num, seed=42,
+                                   angle=rot_angle, speed=0)
+        manual_result = zoom_fn(manual_result, t=t, frame_num=frame_num, seed=42,
+                                 amount=zoom_amount, speed=0)
+        manual_result = np.asarray(manual_result)
+
+        # Fused chain application
+        chain = DeferredEffectChain(
+            ['rotate'],
+            [{'angle': rot_angle, 'speed': 0}],
+            jax_frame, t, frame_num
+        )
+        chain = chain.extend('zoom', {'amount': zoom_amount, 'speed': 0})
+
+        fused_result = np.asarray(interp._force_deferred(chain))
+
+        diff = np.abs(manual_result.astype(float) - fused_result.astype(float))
+        mean_diff = np.mean(diff)
+
+        assert mean_diff < 1.0, f"Manual vs fused: mean diff {mean_diff:.2f}"
+
+        # Check specific pixels
+        h, w = test_img.shape[:2]
+        for y in [0, h//2, h-1]:
+            for x in [0, w//2, w-1]:
+                manual_val = manual_result[y, x]
+                fused_val = fused_result[y, x]
+                pixel_diff = np.abs(manual_val.astype(float) - fused_val.astype(float)).max()
+                assert pixel_diff < 2.0, f"Pixel ({y},{x}): manual={manual_val}, fused={fused_val}"
+
+    def test_chain_with_shrink_zoom_boundary(self, test_env):
+        """Test that shrinking zoom handles boundaries correctly in chain."""
+        import jax.numpy as jnp
+        from streaming.stream_sexp_generic import StreamInterpreter, DeferredEffectChain
+
+        sexp_content = '''(stream "test"
+  :width 128
+  :height 96
+  :seed 42
+
+  (effect rotate :path "../sexp_effects/effects/rotate.sexp")
+  (effect zoom :path "../sexp_effects/effects/zoom.sexp")
+
+  (frame frame)
+)
+'''
+        sexp_path = create_sexp_file(test_env['dir'], sexp_content)
+
+        core_mod.set_random_seed(42)
+        interp = StreamInterpreter(sexp_path, use_jax=True)
+        interp._init()
+
+        test_img = test_env['test_img']
+        jax_frame = jnp.array(test_img)
+        t = 0.5
+        frame_num = 5
+
+        # Parameters that shrink the image (zoom < 1.0)
+        rot_angle = -4.555
+        zoom_amount = 0.9494  # This pulls in from edges, exposing boundaries
+
+        # Manual application
+        rotate_fn = interp.jax_effects['rotate']
+        zoom_fn = interp.jax_effects['zoom']
+
+        manual_result = rotate_fn(jax_frame, t=t, frame_num=frame_num, seed=42,
+                                   angle=rot_angle, speed=0)
+        manual_result = zoom_fn(manual_result, t=t, frame_num=frame_num, seed=42,
+                                 amount=zoom_amount, speed=0)
+        manual_result = np.asarray(manual_result)
+
+        # Fused chain
+        chain = DeferredEffectChain(
+            ['rotate'],
+            [{'angle': rot_angle, 'speed': 0}],
+            jax_frame, t, frame_num
+        )
+        chain = chain.extend('zoom', {'amount': zoom_amount, 'speed': 0})
+
+        fused_result = np.asarray(interp._force_deferred(chain))
+
+        # Check top edge specifically - this is where boundary issues manifest
+        top_edge_manual = manual_result[0, :]
+        top_edge_fused = fused_result[0, :]
+
+        edge_diff = np.abs(top_edge_manual.astype(float) - top_edge_fused.astype(float))
+        mean_edge_diff = np.mean(edge_diff)
+
+        assert mean_edge_diff < 2.0, f"Top edge diff: {mean_edge_diff:.2f}"
+
+        # Check for zeros at edges that shouldn't be there
+        manual_edge_sum = np.sum(top_edge_manual)
+        fused_edge_sum = np.sum(top_edge_fused)
+
+        if manual_edge_sum > 100:  # If manual has significant values
+            assert fused_edge_sum > manual_edge_sum * 0.5, \
+                f"Fused has too many zeros: manual sum={manual_edge_sum}, fused sum={fused_edge_sum}"
+
+
+if __name__ == '__main__':
+    pytest.main([__file__, '-v'])

tests/test_jax_primitives.py

@@ -0,0 +1,334 @@
+#!/usr/bin/env python3
+"""
+Test framework to verify JAX primitives match Python primitives.
+
+Compares output of each primitive through:
+1. Python/NumPy path
+2. JAX path (CPU)
+3. JAX path (GPU) - if available
+
+Reports any mismatches with detailed diffs.
+"""
+import sys
+sys.path.insert(0, '/home/giles/art/art-celery')
+
+import numpy as np
+import json
+from pathlib import Path
+from typing import Dict, List, Tuple, Any, Optional
+from dataclasses import dataclass, field
+
+# Test configuration
+TEST_WIDTH = 64
+TEST_HEIGHT = 48
+TOLERANCE_MEAN = 1.0  # Max allowed mean difference
+TOLERANCE_MAX = 10.0  # Max allowed single pixel difference
+TOLERANCE_PCT = 0.95  # Min % of pixels within ±1
+
+
+@dataclass
+class TestResult:
+    primitive: str
+    passed: bool
+    python_mean: float = 0.0
+    jax_mean: float = 0.0
+    diff_mean: float = 0.0
+    diff_max: float = 0.0
+    pct_within_1: float = 0.0
+    error: str = ""
+
+
+def create_test_frame(w=TEST_WIDTH, h=TEST_HEIGHT, pattern='gradient'):
+    """Create a test frame with known pattern."""
+    if pattern == 'gradient':
+        # Diagonal gradient
+        y, x = np.mgrid[0:h, 0:w]
+        r = (x * 255 / w).astype(np.uint8)
+        g = (y * 255 / h).astype(np.uint8)
+        b = ((x + y) * 127 / (w + h)).astype(np.uint8)
+        return np.stack([r, g, b], axis=2)
+    elif pattern == 'checkerboard':
+        y, x = np.mgrid[0:h, 0:w]
+        check = ((x // 8) + (y // 8)) % 2
+        v = (check * 255).astype(np.uint8)
+        return np.stack([v, v, v], axis=2)
+    elif pattern == 'solid':
+        return np.full((h, w, 3), 128, dtype=np.uint8)
+    else:
+        return np.random.randint(0, 255, (h, w, 3), dtype=np.uint8)
+
+
+def compare_outputs(py_out, jax_out) -> Tuple[float, float, float]:
+    """Compare two outputs, return (mean_diff, max_diff, pct_within_1)."""
+    if py_out is None or jax_out is None:
+        return float('inf'), float('inf'), 0.0
+
+    if isinstance(py_out, dict) and isinstance(jax_out, dict):
+        # Compare coordinate maps
+        diffs = []
+        for k in py_out:
+            if k in jax_out:
+                py_arr = np.asarray(py_out[k])
+                jax_arr = np.asarray(jax_out[k])
+                if py_arr.shape == jax_arr.shape:
+                    diff = np.abs(py_arr.astype(float) - jax_arr.astype(float))
+                    diffs.append(diff)
+        if diffs:
+            all_diff = np.concatenate([d.flatten() for d in diffs])
+            return float(np.mean(all_diff)), float(np.max(all_diff)), float(np.mean(all_diff <= 1))
+        return float('inf'), float('inf'), 0.0
+
+    py_arr = np.asarray(py_out)
+    jax_arr = np.asarray(jax_out)
+
+    if py_arr.shape != jax_arr.shape:
+        return float('inf'), float('inf'), 0.0
+
+    diff = np.abs(py_arr.astype(float) - jax_arr.astype(float))
+    return float(np.mean(diff)), float(np.max(diff)), float(np.mean(diff <= 1))
+
+
+# ============================================================================
+# Primitive Test Definitions
+# ============================================================================
+
+PRIMITIVE_TESTS = {
+    # Geometry primitives
+    'geometry:ripple-displace': {
+        'args': [TEST_WIDTH, TEST_HEIGHT, 5, 10, TEST_WIDTH/2, TEST_HEIGHT/2, 1, 0.5],
+        'returns': 'coords',
+    },
+    'geometry:rotate-img': {
+        'args': ['frame', 45],
+        'returns': 'frame',
+    },
+    'geometry:scale-img': {
+        'args': ['frame', 1.5],
+        'returns': 'frame',
+    },
+    'geometry:flip-h': {
+        'args': ['frame'],
+        'returns': 'frame',
+    },
+    'geometry:flip-v': {
+        'args': ['frame'],
+        'returns': 'frame',
+    },
+
+    # Color operations
+    'color_ops:invert': {
+        'args': ['frame'],
+        'returns': 'frame',
+    },
+    'color_ops:grayscale': {
+        'args': ['frame'],
+        'returns': 'frame',
+    },
+    'color_ops:brightness': {
+        'args': ['frame', 1.5],
+        'returns': 'frame',
+    },
+    'color_ops:contrast': {
+        'args': ['frame', 1.5],
+        'returns': 'frame',
+    },
+    'color_ops:hue-shift': {
+        'args': ['frame', 90],
+        'returns': 'frame',
+    },
+
+    # Image operations
+    'image:width': {
+        'args': ['frame'],
+        'returns': 'scalar',
+    },
+    'image:height': {
+        'args': ['frame'],
+        'returns': 'scalar',
+    },
+    'image:channel': {
+        'args': ['frame', 0],
+        'returns': 'array',
+    },
+
+    # Blending
+    'blending:blend': {
+        'args': ['frame', 'frame2', 0.5],
+        'returns': 'frame',
+    },
+    'blending:blend-add': {
+        'args': ['frame', 'frame2'],
+        'returns': 'frame',
+    },
+    'blending:blend-multiply': {
+        'args': ['frame', 'frame2'],
+        'returns': 'frame',
+    },
+}
+
+
+def run_python_primitive(interp, prim_name: str, test_def: dict, frame: np.ndarray, frame2: np.ndarray) -> Any:
+    """Run a primitive through the Python interpreter."""
+    if prim_name not in interp.primitives:
+        return None
+
+    func = interp.primitives[prim_name]
+    args = []
+    for a in test_def['args']:
+        if a == 'frame':
+            args.append(frame.copy())
+        elif a == 'frame2':
+            args.append(frame2.copy())
+        else:
+            args.append(a)
+
+    try:
+        return func(*args)
+    except Exception as e:
+        return None
+
+
+def run_jax_primitive(prim_name: str, test_def: dict, frame: np.ndarray, frame2: np.ndarray) -> Any:
+    """Run a primitive through the JAX compiler."""
+    try:
+        from streaming.sexp_to_jax import JaxCompiler
+        import jax.numpy as jnp
+
+        compiler = JaxCompiler()
+
+        # Build a simple expression to test the primitive
+        from sexp_effects.parser import Symbol, Keyword
+
+        args = []
+        env = {'frame': jnp.array(frame), 'frame2': jnp.array(frame2)}
+
+        for a in test_def['args']:
+            if a == 'frame':
+                args.append(Symbol('frame'))
+            elif a == 'frame2':
+                args.append(Symbol('frame2'))
+            else:
+                args.append(a)
+
+        # Create expression: (prim_name arg1 arg2 ...)
+        expr = [Symbol(prim_name)] + args
+
+        result = compiler._eval(expr, env)
+
+        if hasattr(result, '__array__'):
+            return np.asarray(result)
+        return result
+
+    except Exception as e:
+        return None
+
+
+def test_primitive(interp, prim_name: str, test_def: dict) -> TestResult:
+    """Test a single primitive."""
+    frame = create_test_frame(pattern='gradient')
+    frame2 = create_test_frame(pattern='checkerboard')
+
+    result = TestResult(primitive=prim_name, passed=False)
+
+    # Run Python version
+    try:
+        py_out = run_python_primitive(interp, prim_name, test_def, frame, frame2)
+        if py_out is not None and hasattr(py_out, 'shape'):
+            result.python_mean = float(np.mean(py_out))
+    except Exception as e:
+        result.error = f"Python error: {e}"
+        return result
+
+    # Run JAX version
+    try:
+        jax_out = run_jax_primitive(prim_name, test_def, frame, frame2)
+        if jax_out is not None and hasattr(jax_out, 'shape'):
+            result.jax_mean = float(np.mean(jax_out))
+    except Exception as e:
+        result.error = f"JAX error: {e}"
+        return result
+
+    if py_out is None:
+        result.error = "Python returned None"
+        return result
+    if jax_out is None:
+        result.error = "JAX returned None"
+        return result
+
+    # Compare
+    diff_mean, diff_max, pct = compare_outputs(py_out, jax_out)
+    result.diff_mean = diff_mean
+    result.diff_max = diff_max
+    result.pct_within_1 = pct
+
+    # Check pass/fail
+    result.passed = (
+        diff_mean <= TOLERANCE_MEAN and
+        diff_max <= TOLERANCE_MAX and
+        pct >= TOLERANCE_PCT
+    )
+
+    if not result.passed:
+        result.error = f"Diff too large: mean={diff_mean:.2f}, max={diff_max:.1f}, pct={pct:.1%}"
+
+    return result
+
+
+def discover_primitives(interp) -> List[str]:
+    """Discover all primitives available in the interpreter."""
+    return sorted(interp.primitives.keys())
+
+
+def run_all_tests(verbose=True):
+    """Run all primitive tests."""
+    import warnings
+    warnings.filterwarnings('ignore')
+
+    import os
+    os.chdir('/home/giles/art/test')
+
+    from streaming.stream_sexp_generic import StreamInterpreter
+    from sexp_effects.primitive_libs import core as core_mod
+
+    core_mod.set_random_seed(42)
+
+    # Create interpreter to get Python primitives
+    interp = StreamInterpreter('effects/quick_test_explicit.sexp', use_jax=False)
+    interp._init()
+
+    results = []
+
+    print("=" * 60)
+    print("JAX PRIMITIVE TEST SUITE")
+    print("=" * 60)
+
+    # Test defined primitives
+    for prim_name, test_def in PRIMITIVE_TESTS.items():
+        result = test_primitive(interp, prim_name, test_def)
+        results.append(result)
+
+        status = "✓ PASS" if result.passed else "✗ FAIL"
+        if verbose:
+            print(f"{status} {prim_name}")
+            if not result.passed:
+                print(f"       {result.error}")
+
+    # Summary
+    passed = sum(1 for r in results if r.passed)
+    failed = sum(1 for r in results if not r.passed)
+
+    print("\n" + "=" * 60)
+    print(f"SUMMARY: {passed} passed, {failed} failed")
+    print("=" * 60)
+
+    if failed > 0:
+        print("\nFailed primitives:")
+        for r in results:
+            if not r.passed:
+                print(f"  - {r.primitive}: {r.error}")
+
+    return results
+
+
+if __name__ == '__main__':
+    run_all_tests()

tests/test_xector.py

@@ -0,0 +1,305 @@
+"""
+Tests for xector primitives - parallel array operations.
+"""
+
+import pytest
+import numpy as np
+from sexp_effects.primitive_libs.xector import (
+    Xector,
+    xector_red, xector_green, xector_blue, xector_rgb,
+    xector_x_coords, xector_y_coords, xector_x_norm, xector_y_norm,
+    xector_dist_from_center,
+    alpha_add, alpha_sub, alpha_mul, alpha_div, alpha_sqrt, alpha_clamp,
+    alpha_sin, alpha_cos, alpha_sq,
+    alpha_lt, alpha_gt, alpha_eq,
+    beta_add, beta_mul, beta_min, beta_max, beta_mean, beta_count,
+    xector_where, xector_fill, xector_zeros, xector_ones,
+    is_xector,
+)
+
+
+class TestXectorBasics:
+    """Test Xector class basic operations."""
+
+    def test_create_from_list(self):
+        x = Xector([1, 2, 3])
+        assert len(x) == 3
+        assert is_xector(x)
+
+    def test_create_from_numpy(self):
+        arr = np.array([1.0, 2.0, 3.0])
+        x = Xector(arr)
+        assert len(x) == 3
+        np.testing.assert_array_equal(x.to_numpy(), arr.astype(np.float32))
+
+    def test_implicit_add(self):
+        a = Xector([1, 2, 3])
+        b = Xector([4, 5, 6])
+        c = a + b
+        np.testing.assert_array_equal(c.to_numpy(), [5, 7, 9])
+
+    def test_implicit_mul(self):
+        a = Xector([1, 2, 3])
+        b = Xector([2, 2, 2])
+        c = a * b
+        np.testing.assert_array_equal(c.to_numpy(), [2, 4, 6])
+
+    def test_scalar_broadcast(self):
+        a = Xector([1, 2, 3])
+        c = a + 10
+        np.testing.assert_array_equal(c.to_numpy(), [11, 12, 13])
+
+    def test_scalar_broadcast_rmul(self):
+        a = Xector([1, 2, 3])
+        c = 2 * a
+        np.testing.assert_array_equal(c.to_numpy(), [2, 4, 6])
+
+
+class TestAlphaOperations:
+    """Test α (element-wise) operations."""
+
+    def test_alpha_add(self):
+        a = Xector([1, 2, 3])
+        b = Xector([4, 5, 6])
+        c = alpha_add(a, b)
+        np.testing.assert_array_equal(c.to_numpy(), [5, 7, 9])
+
+    def test_alpha_add_multi(self):
+        a = Xector([1, 2, 3])
+        b = Xector([1, 1, 1])
+        c = Xector([10, 10, 10])
+        d = alpha_add(a, b, c)
+        np.testing.assert_array_equal(d.to_numpy(), [12, 13, 14])
+
+    def test_alpha_mul_scalar(self):
+        a = Xector([1, 2, 3])
+        c = alpha_mul(a, 2)
+        np.testing.assert_array_equal(c.to_numpy(), [2, 4, 6])
+
+    def test_alpha_sqrt(self):
+        a = Xector([1, 4, 9, 16])
+        c = alpha_sqrt(a)
+        np.testing.assert_array_equal(c.to_numpy(), [1, 2, 3, 4])
+
+    def test_alpha_clamp(self):
+        a = Xector([-5, 0, 5, 10, 15])
+        c = alpha_clamp(a, 0, 10)
+        np.testing.assert_array_equal(c.to_numpy(), [0, 0, 5, 10, 10])
+
+    def test_alpha_sin_cos(self):
+        a = Xector([0, np.pi/2, np.pi])
+        s = alpha_sin(a)
+        c = alpha_cos(a)
+        np.testing.assert_array_almost_equal(s.to_numpy(), [0, 1, 0], decimal=5)
+        np.testing.assert_array_almost_equal(c.to_numpy(), [1, 0, -1], decimal=5)
+
+    def test_alpha_sq(self):
+        a = Xector([1, 2, 3, 4])
+        c = alpha_sq(a)
+        np.testing.assert_array_equal(c.to_numpy(), [1, 4, 9, 16])
+
+    def test_alpha_comparison(self):
+        a = Xector([1, 2, 3, 4])
+        b = Xector([2, 2, 2, 2])
+        lt = alpha_lt(a, b)
+        gt = alpha_gt(a, b)
+        eq = alpha_eq(a, b)
+        np.testing.assert_array_equal(lt.to_numpy(), [True, False, False, False])
+        np.testing.assert_array_equal(gt.to_numpy(), [False, False, True, True])
+        np.testing.assert_array_equal(eq.to_numpy(), [False, True, False, False])
+
+
+class TestBetaOperations:
+    """Test β (reduction) operations."""
+
+    def test_beta_add(self):
+        a = Xector([1, 2, 3, 4])
+        assert beta_add(a) == 10
+
+    def test_beta_mul(self):
+        a = Xector([1, 2, 3, 4])
+        assert beta_mul(a) == 24
+
+    def test_beta_min_max(self):
+        a = Xector([3, 1, 4, 1, 5, 9, 2, 6])
+        assert beta_min(a) == 1
+        assert beta_max(a) == 9
+
+    def test_beta_mean(self):
+        a = Xector([1, 2, 3, 4])
+        assert beta_mean(a) == 2.5
+
+    def test_beta_count(self):
+        a = Xector([1, 2, 3, 4, 5])
+        assert beta_count(a) == 5
+
+
+class TestFrameConversion:
+    """Test frame/xector conversion."""
+
+    def test_extract_channels(self):
+        # Create a 2x2 RGB frame
+        frame = np.array([
+            [[255, 0, 0], [0, 255, 0]],
+            [[0, 0, 255], [128, 128, 128]]
+        ], dtype=np.uint8)
+
+        r = xector_red(frame)
+        g = xector_green(frame)
+        b = xector_blue(frame)
+
+        assert len(r) == 4
+        np.testing.assert_array_equal(r.to_numpy(), [255, 0, 0, 128])
+        np.testing.assert_array_equal(g.to_numpy(), [0, 255, 0, 128])
+        np.testing.assert_array_equal(b.to_numpy(), [0, 0, 255, 128])
+
+    def test_rgb_roundtrip(self):
+        # Create a 2x2 RGB frame
+        frame = np.array([
+            [[100, 150, 200], [50, 75, 100]],
+            [[200, 100, 50], [25, 50, 75]]
+        ], dtype=np.uint8)
+
+        r = xector_red(frame)
+        g = xector_green(frame)
+        b = xector_blue(frame)
+
+        reconstructed = xector_rgb(r, g, b)
+        np.testing.assert_array_equal(reconstructed, frame)
+
+    def test_modify_and_reconstruct(self):
+        frame = np.array([
+            [[100, 100, 100], [100, 100, 100]],
+            [[100, 100, 100], [100, 100, 100]]
+        ], dtype=np.uint8)
+
+        r = xector_red(frame)
+        g = xector_green(frame)
+        b = xector_blue(frame)
+
+        # Double red channel
+        r_doubled = r * 2
+
+        result = xector_rgb(r_doubled, g, b)
+
+        # Red should be 200, others unchanged
+        assert result[0, 0, 0] == 200
+        assert result[0, 0, 1] == 100
+        assert result[0, 0, 2] == 100
+
+
+class TestCoordinates:
+    """Test coordinate generation."""
+
+    def test_x_coords(self):
+        frame = np.zeros((2, 3, 3), dtype=np.uint8)  # 2 rows, 3 cols
+        x = xector_x_coords(frame)
+        # Should be [0,1,2, 0,1,2] (x coords repeated for each row)
+        np.testing.assert_array_equal(x.to_numpy(), [0, 1, 2, 0, 1, 2])
+
+    def test_y_coords(self):
+        frame = np.zeros((2, 3, 3), dtype=np.uint8)  # 2 rows, 3 cols
+        y = xector_y_coords(frame)
+        # Should be [0,0,0, 1,1,1] (y coords for each pixel)
+        np.testing.assert_array_equal(y.to_numpy(), [0, 0, 0, 1, 1, 1])
+
+    def test_normalized_coords(self):
+        frame = np.zeros((2, 3, 3), dtype=np.uint8)
+        x = xector_x_norm(frame)
+        y = xector_y_norm(frame)
+
+        # x should go 0 to 1 across width
+        assert x.to_numpy()[0] == 0
+        assert x.to_numpy()[2] == 1
+
+        # y should go 0 to 1 down height
+        assert y.to_numpy()[0] == 0
+        assert y.to_numpy()[3] == 1
+
+
+class TestConditional:
+    """Test conditional operations."""
+
+    def test_where(self):
+        cond = Xector([True, False, True, False])
+        true_val = Xector([1, 1, 1, 1])
+        false_val = Xector([0, 0, 0, 0])
+
+        result = xector_where(cond, true_val, false_val)
+        np.testing.assert_array_equal(result.to_numpy(), [1, 0, 1, 0])
+
+    def test_where_with_comparison(self):
+        a = Xector([1, 5, 3, 7])
+        threshold = 4
+
+        # Elements > 4 become 255, others become 0
+        result = xector_where(alpha_gt(a, threshold), 255, 0)
+        np.testing.assert_array_equal(result.to_numpy(), [0, 255, 0, 255])
+
+    def test_fill(self):
+        frame = np.zeros((2, 3, 3), dtype=np.uint8)
+        x = xector_fill(42, frame)
+        assert len(x) == 6
+        assert all(v == 42 for v in x.to_numpy())
+
+    def test_zeros_ones(self):
+        frame = np.zeros((2, 2, 3), dtype=np.uint8)
+        z = xector_zeros(frame)
+        o = xector_ones(frame)
+
+        assert all(v == 0 for v in z.to_numpy())
+        assert all(v == 1 for v in o.to_numpy())
+
+
+class TestInterpreterIntegration:
+    """Test xector operations through the interpreter."""
+
+    def test_xector_vignette_effect(self):
+        from sexp_effects.interpreter import Interpreter
+
+        interp = Interpreter(minimal_primitives=True)
+
+        # Load the xector vignette effect
+        interp.load_effect('sexp_effects/effects/xector_vignette.sexp')
+
+        # Create a test frame (white)
+        frame = np.full((100, 100, 3), 255, dtype=np.uint8)
+
+        # Run effect
+        result, state = interp.run_effect('xector_vignette', frame, {'strength': 0.5}, {})
+
+        # Center should be brighter than corners
+        center = result[50, 50]
+        corner = result[0, 0]
+
+        assert center.mean() > corner.mean(), "Center should be brighter than corners"
+        # Corners should be darkened
+        assert corner.mean() < 255, "Corners should be darkened"
+
+    def test_implicit_elementwise(self):
+        """Test that regular + works element-wise on xectors."""
+        from sexp_effects.interpreter import Interpreter
+
+        interp = Interpreter(minimal_primitives=True)
+        # Load xector primitives
+        from sexp_effects.primitive_libs.xector import PRIMITIVES
+        for name, fn in PRIMITIVES.items():
+            interp.global_env.set(name, fn)
+
+        # Parse and eval a simple xector expression
+        from sexp_effects.parser import parse
+        expr = parse('(+ (red frame) 10)')
+
+        # Create test frame
+        frame = np.full((2, 2, 3), 100, dtype=np.uint8)
+        interp.global_env.set('frame', frame)
+
+        result = interp.eval(expr)
+
+        # Should be a xector with values 110
+        assert is_xector(result)
+        np.testing.assert_array_equal(result.to_numpy(), [110, 110, 110, 110])
+
+
+if __name__ == '__main__':
+    pytest.main([__file__, '-v'])