198 lines
5.9 KiB
Markdown
198 lines
5.9 KiB
Markdown
# Art DAG Architecture
|
|
|
|
## Overview
|
|
|
|
Art DAG is a content-addressed creative system with federated ownership. It separates rendering (L1) from social/ownership (L2), enabling distributed creative work with automatic attribution and revenue sharing.
|
|
|
|
## Two-Layer Architecture
|
|
|
|
### Level 1 - Rendering Layer
|
|
|
|
**L1 Server** (Celery workers)
|
|
- Executes rendering jobs
|
|
- Records provenance in the DAG
|
|
- Returns content-addressed outputs (SHA3-256)
|
|
- Stateless, horizontally scalable
|
|
- Most renders are junk/experiments - that's fine
|
|
|
|
**L1 Client** (AI-assisted front-end)
|
|
- "Make me a video that rotates" → AI designs config
|
|
- Sends configs + assets to L1 server
|
|
- Receives rendered output with provenance
|
|
- Decides what's worth keeping
|
|
|
|
### Level 2 - Social/Federation Layer
|
|
|
|
**L2 Server** (ActivityPub)
|
|
- Ownership registry: "I made this with X, Y, Z"
|
|
- Federation: follow/followers across the fediverse
|
|
- Web UI: creators share their work
|
|
- Revenue distribution: ads flow down provenance chain
|
|
- Only shared content is persistent (curated from L1 junk)
|
|
|
|
**Key insight**: L1 produces lots of experimental renders. L2 is where the creator decides "this is the good stuff" and shares it via ActivityPub. Persistence = intentional sharing.
|
|
|
|
## Data Flow
|
|
|
|
```
|
|
User prompt: "make me a cat video"
|
|
│
|
|
▼
|
|
┌─────────────┐
|
|
│ AI Client │ Designs config from natural language
|
|
└─────────────┘
|
|
│
|
|
▼
|
|
┌─────────────┐
|
|
│ L1 Server │ Renders, records provenance in DAG
|
|
└─────────────┘
|
|
│
|
|
▼
|
|
User reviews output
|
|
│
|
|
▼ (if good)
|
|
┌─────────────┐
|
|
│ L2 Server │ Shares via ActivityPub, registers ownership
|
|
└─────────────┘
|
|
│
|
|
▼
|
|
Federated web (followers see it)
|
|
```
|
|
|
|
## Effects and Recipes
|
|
|
|
**Effects** are atomic transformations:
|
|
- `effect:identity` - returns input unchanged
|
|
- `effect:dog` - ignores input, returns dog.mkv
|
|
- `effect:glitch` - applies glitch effect
|
|
- Registered with content hash, owned by creator
|
|
|
|
**Recipes** are compositions of effects:
|
|
- A single effect IS a recipe (the simplest case)
|
|
- Complex recipes chain/branch multiple effects into a DAG
|
|
- Registered with content hash, owned by creator
|
|
- Recipe composer earns when their recipe is used
|
|
|
|
```
|
|
Simple recipe: input → effect:dog → output
|
|
|
|
Complex recipe: input ─┬→ effect:resize ──┬→ effect:overlay → output
|
|
└→ effect:glitch ──┘
|
|
```
|
|
|
|
Both effects and recipes are registered assets. Both creators earn from downstream usage.
|
|
|
|
## Provenance Chain
|
|
|
|
Every output knows its lineage:
|
|
|
|
```json
|
|
{
|
|
"output": {"name": "cool-video", "content_hash": "abc123..."},
|
|
"inputs": [
|
|
{"name": "cat-footage", "content_hash": "def456..."}
|
|
],
|
|
"effects": [
|
|
{"name": "effect:glitch", "content_hash": "ghi789..."}
|
|
],
|
|
"infrastructure": {
|
|
"software": {"name": "infra:artdag", "content_hash": "..."},
|
|
"hardware": {"name": "infra:render-farm", "content_hash": "..."}
|
|
}
|
|
}
|
|
```
|
|
|
|
## Revenue Model
|
|
|
|
### How money enters
|
|
- Advertising on L2 server web interface
|
|
- Premium features (faster rendering, more storage)
|
|
- Direct payments for outputs (commissions, licensing)
|
|
- **Storage fees** to retain ownership:
|
|
- Free tier: platform gets rights, can publish to YouTube/monetize elsewhere
|
|
- Paid tier: you keep full ownership
|
|
- Honest trade, not hidden in ToS
|
|
- **Render quality fees** (L1 server):
|
|
- Free: watermarks, limited resolution, limited length
|
|
- Paid: clean output, full resolution, no limits
|
|
|
|
### How money flows
|
|
Revenue splits down the provenance chain:
|
|
|
|
```
|
|
Ad revenue from video view
|
|
│
|
|
├── 40% → footage creator (input)
|
|
├── 25% → effect author
|
|
├── 20% → recipe/config designer
|
|
└── 15% → L1 render infrastructure
|
|
```
|
|
|
|
Exact splits could be configurable per-output or protocol-defined.
|
|
|
|
### Trust and Federation
|
|
|
|
**Honourable L2 servers:**
|
|
- Distribute revenue fairly down provenance chain
|
|
- Respect attribution
|
|
- Federate openly
|
|
|
|
**Dishonourable L2 servers:**
|
|
- Keep revenue for themselves
|
|
- Strip attribution
|
|
- Get "sent to Coventry" - defederated by peers
|
|
|
|
Self-policing network like Mastodon instance blocking. Reputation matters.
|
|
|
|
## Proof of Creativity
|
|
|
|
Unlike Proof of Work (burn electricity) or Proof of Stake (lock capital):
|
|
|
|
**Proof of Creativity** = your contribution is measured by downstream usage
|
|
|
|
- Make useful footage → gets used → you earn
|
|
- Write clever effects → gets used → you earn
|
|
- Design good recipes → gets used → you earn
|
|
|
|
Not artificial scarcity. Not speculation. Actual creative value.
|
|
|
|
## Why share here vs Facebook/YouTube?
|
|
|
|
**Facebook/YouTube:**
|
|
- You upload → they own it
|
|
- Their algorithm decides visibility
|
|
- They monetize, you get scraps
|
|
- They can delete/demonetize anytime
|
|
- Your audience belongs to them
|
|
|
|
**Art DAG (federated):**
|
|
- You publish → you own it (cryptographically signed)
|
|
- Your server, your algorithm
|
|
- Provenance ensures fair payment
|
|
- Nobody can delete your work
|
|
- Your audience connects directly to you
|
|
|
|
## Repository Structure
|
|
|
|
```
|
|
GitHub (gilesbradshaw/art-dag)
|
|
└── Core primitives: DAG, signing, ActivityPub, effects
|
|
|
|
git.rose-ash.com/art-dag/
|
|
├── registry → Asset registry + signatures
|
|
├── effects → Effect implementations (dog, etc.)
|
|
├── recipes → Render scripts
|
|
├── art → Rendered outputs
|
|
├── celery → L1 distributed rendering
|
|
└── document → This documentation
|
|
```
|
|
|
|
## Technical Foundation
|
|
|
|
- **Hashing**: SHA3-256 (quantum-resistant, full 64 char)
|
|
- **Signing**: RSA-2048 with RsaSignature2017
|
|
- **Federation**: ActivityPub protocol
|
|
- **Identity**: @user@domain.com format
|
|
- **URLs**: Immutable via git commit hashes
|
|
- **Rendering**: Celery + Redis for distributed jobs
|