transcendedhacker
0f0f088ab2
Introduces a production-ready, schema-driven prompt generation engine for ComfyUI, including a new 'everyday_casual' library with 290+ atomic options. Major updates include a comprehensive README, scaling and library comparison docs, a strict LICENSE (CC BY-NC 4.0), and a refactored __init__.py with improved node structure, negative prompt support, and detailed docstrings. The engine (engine/composer.py) now enforces strict separation of concerns, supports addons, intent flags, and deterministic assembly. Documentation and changelog updated to reflect the new scalable, maintainable architecture.
138 lines
4.1 KiB
Markdown
138 lines
4.1 KiB
Markdown
# Development Notes
|
||
|
||
## v2.0 - Production Architecture
|
||
|
||
Schema-driven prompt generation engine with 300+ atomic compositional elements across 17 structural blocks.
|
||
|
||
## System Architecture
|
||
|
||
**Separation of Concerns**:
|
||
- Engine: Deterministic assembly logic (no domain knowledge)
|
||
- Schema: Structural contract (ordering, weighting, rules)
|
||
- Libraries: Content datasets (pure options, no logic)
|
||
|
||
**Data Flow**:
|
||
User selections → Schema validation → Library lookup → Weight application → Token ordering → Assembled prompt
|
||
|
||
## Block Organization
|
||
|
||
Schema defines 17 compositional blocks in priority order:
|
||
|
||
subject → age → hair_color → hair_style → hair_ornament → outfit → outfit_detail → pose → mood → location → location_detail → atmosphere → lighting → angle → frame → filter → technical
|
||
|
||
## None Handling
|
||
|
||
Selecting "None" in any dropdown:
|
||
- Block completely omitted from assembly
|
||
- No blank text inserted
|
||
- Clean composition maintained
|
||
|
||
Example:
|
||
- hair_ornament: None → No ornament text appears
|
||
- outfit_detail: None → Just outfit, no fabric details
|
||
|
||
## Alphabetical Sorting
|
||
|
||
All dropdowns alphabetically sorted for predictable navigation.
|
||
"None" always first regardless of sorting.
|
||
|
||
Benefits:
|
||
- Scannable even with 100+ options per block
|
||
- No search interface needed
|
||
- Consistent muscle memory for frequent options
|
||
|
||
## Current Scale
|
||
|
||
17 blocks × ~18 average options = ~300 total selections
|
||
|
||
Proven capacity: System handles 1000+ options without performance degradation.
|
||
|
||
## Performance Characteristics
|
||
|
||
v1 (sentence-based): ~300s generation time
|
||
v2 (atomic composition): <100s generation time
|
||
|
||
Clean, structured prompts encode faster and produce more consistent results.
|
||
|
||
## Scaling Strategies
|
||
|
||
### Vertical Scaling (expand existing blocks)
|
||
Add options to current blocks from reference documentation:
|
||
- Parse HairColors.txt → add all 40+ colors
|
||
- Parse AngleShot.txt → add all 100+ camera angles
|
||
- Parse SexyPoses.txt → add all 200+ pose variations
|
||
- Parse LocationsIndoor.txt → add all 300+ interior settings
|
||
- Parse Atmospheres.txt → add all 100+ atmospheric conditions
|
||
|
||
### Horizontal Scaling (add new blocks)
|
||
Introduce additional compositional elements as needed:
|
||
- Facial expression control
|
||
- Hand positioning
|
||
- Environmental props
|
||
- Weather conditions
|
||
- Color palette specifications
|
||
|
||
### Library Specialization
|
||
Create context-specific libraries for distinct use cases:
|
||
- portrait_casual.json
|
||
- portrait_formal.json
|
||
- portrait_fantasy.json
|
||
- portrait_technical.json
|
||
|
||
## Technical Constraints
|
||
|
||
**ComfyUI Limitation**: Dropdowns cannot dynamically update based on other selections.
|
||
|
||
Cannot implement:
|
||
- "outfit: dress" → auto-filter outfit_detail to dress-specific options
|
||
- "location: indoor" → auto-filter location_detail to indoor conditions
|
||
|
||
Workarounds:
|
||
1. Comprehensive libraries covering all valid combinations
|
||
2. Specialized libraries for specific contexts
|
||
3. Schema-level validation (future implementation)
|
||
|
||
## Extension Workflow
|
||
|
||
To add options:
|
||
|
||
1. Edit target library JSON file
|
||
2. Add entries in appropriate block
|
||
3. Follow format: `"Option Name": "descriptive prompt text"`
|
||
4. Save file
|
||
5. Restart ComfyUI
|
||
|
||
No engine modification required. Alphabetical sorting automatic.
|
||
|
||
## Intent Flags (Planned)
|
||
|
||
Schema defines intent flags for future implementation:
|
||
- exposure_level: [covered, implied, nude]
|
||
- camera_relationship: [facing, candid, over_shoulder, pov, back_view]
|
||
|
||
Default values ensure safe, predictable behavior.
|
||
Explicit opt-in required for non-default states.
|
||
|
||
## Addon Support (Planned)
|
||
|
||
Each block will support local addon text:
|
||
- Modifies only that block's output
|
||
- Merges inline via comma separator
|
||
- Example: "woman" + addon "athletic build" → "woman, athletic build"
|
||
|
||
## Provenance Tracking
|
||
|
||
Optional non-rendering comment for attribution:
|
||
```
|
||
# Generated using Mode_personal_node — CC BY-NC 4.0
|
||
```
|
||
|
||
Disabled by default. User-removable. Attribution enforced via licensing, not technical restrictions.
|
||
|
||
## Maintenance Philosophy
|
||
|
||
- Engine code frozen unless structural requirements change
|
||
- Schema evolves only for new architectural features
|
||
- Libraries expand freely without affecting engine
|
||
- Backwards compatibility maintained across minor versions
|