This ComfyUI custom node integrates various advanced diffusion model techniques for consistent character and scene generation, targeting users who need fine-grained control over multi-subject storytelling and personalization in image synthesis. It offers a modular approach to leverage multiple state-of-the-art methods within the ComfyUI ecosystem.

How It Works

The node acts as a wrapper, providing ComfyUI nodes for distinct diffusion techniques like StoryDiffusion, MS-Diffusion, Kolor, Pulid, Flux, and others. It facilitates multi-subject generation, character consistency, and identity preservation by integrating specialized adapter models (e.g., IP-Adapter, PhotoMaker) and ID migration methods. The architecture allows users to combine these techniques, offering flexibility in achieving complex visual narratives.

Quick Start & Requirements

• Installation: Clone the repository into the ComfyUI custom nodes directory: git clone https://github.com/smthemex/ComfyUI_StoryDiffusion.git.
• Dependencies: Install requirements with pip install -r requirements.txt. Additional libraries like insightface may be needed for specific models.
• Models: Requires downloading various model checkpoints (e.g., .safetensors, .bin) and placing them in designated ComfyUI model directories. Specific workflow JSON files are provided for testing.
• Hardware: Some methods, particularly those involving multiple models or higher precision, may require significant VRAM (e.g., 16GB+ recommended, some methods mention 27GB+).

Highlighted Details

• Supports multiple ID migration methods for character consistency.
• Integrates with recent research papers and models like PhotoMaker, IP-Adapter, MS-Diffusion, Kolors, and Flux.
• Offers solutions for dual-character composition and prompt error fixes.
• Includes support for various quantization methods (FP8, GGUF, SVDQuant) for potential performance gains.

Maintenance & Community

The repository is actively updated with new methods and bug fixes. Links to example workflows and citations for integrated research papers are provided.

Licensing & Compatibility

The repository itself does not specify a license. However, it integrates models and techniques that may have their own licenses. Users should verify compatibility for commercial use based on the underlying models and their respective licenses.

Limitations & Caveats

Some newer methods (e.g., RealCustom, InstantCharacter) are noted as being slow on consumer hardware (e.g., RTX 4070 12G). Certain features might require specific versions of dependencies (e.g., optimum-quanto >= 0.2.4). Some models require manual downloading and placement, and the README indicates that some older features have been removed in V2 updates.