- Add 10 ComfyUI nodes for multi-shot video storytelling
- Integrate StoryMem repository via git submodule
- Implement memory-conditioned video generation (T2V, M2V, MI2V, MM2V)
- Add sliding window memory buffer system
- Include comprehensive documentation (README, INTEGRATION, EXAMPLES)
- Add model download script and configuration
- Support VRAM optimization with model offloading
- Provide tensor format conversion utilities
- Include test suite for memory buffer
Features:
- Shot-level node architecture for maximum flexibility
- Memory management with sliding window strategy
- Three generation modes (M2V, MI2V, MM2V)
- Automatic keyframe extraction
- Scene cut detection and handling
- Configurable quality settings
- Fallback mode without full integration
Documentation:
- README.md: User guide with installation and usage
- INTEGRATION.md: Technical integration guide
- EXAMPLES.md: Workflow examples and best practices
- CHANGELOG.md: Version history
Requirements:
- Models: ~40-60GB storage, 80GB VRAM (or 24GB+ with offloading)
- Python 3.10+, PyTorch 2.4.0+, ComfyUI
See CHANGELOG.md for complete details.
Bharat Shetty Barkur
c24152d9ea