ComfyUI_ExtraModels is an extension designed to enhance the capabilities of ComfyUI by adding support for various image diffusion models. This extension allows AI artists to leverage different models to create high-quality images, providing more flexibility and creative options. Whether you are looking to generate photorealistic images, artistic renditions, or specific styles, ComfyUI_ExtraModels offers a range of models to suit your needs.
ComfyUI_ExtraModels integrates multiple image diffusion models into ComfyUI, enabling users to select and use different models for their image generation tasks. Image diffusion models work by iteratively refining an image from a noisy starting point, guided by a text prompt or other conditioning information. This process involves several steps where the model predicts and corrects the noise, gradually improving the image quality and aligning it with the given prompt.
Think of it like sculpting a statue from a block of marble. The initial block is rough and undefined (noisy image), and with each step, the sculptor (model) refines the shape, adding details and removing imperfections until the final artwork (high-quality image) is revealed.
PixArt
- T5 Text Encoder: Uses T5 instead of CLIP for text encoding, supporting longer prompts.
- Resolutions: Available in 512px and 1024px versions.
- Compatibility: Works with SD1.5 VAE.
- Optimization: Requires xformers for better image quality.
PixArt Sigma
- Functionality: Similar to PixArt but optimized for different resolutions (512px, 1024px, 2K).
- VAE Requirement: Requires SDXL VAE for optimal performance.
PixArt LCM
- Model: Uses a specific LCM model for image generation.
- Settings: Requires low CFG (1.1-1.5) and around 5 steps for best results.
HunYuan DiT
- Text Encoders: Uses two text encoders (CLIP and mT5) for better text-to-image alignment.
- Model File: Requires specific model files for operation.
- VAE: Compatible with SDXL VAE.
DiT
- Class Labels: Uses class labels instead of text prompts.
- Image Sizes: Limited to 256x256 or 512x512 images.
- Optimization: Works in FP16 but lacks other optimizations.
T5
- Model Files: Requires specific T5 model files for text encoding.
- Memory Usage: Can use significant system RAM or VRAM depending on configuration.
MiaoBi
- Language: A Chinese text-to-image model.
- Compatibility: Compatible with mainstream plugins like Lora, Controlnet, T2I Adapter.
VAE
- Custom VAEs: Supports loading custom VAE models for different image generation tasks.
- Types: Includes models like Consistency Decoder and Deflickering Decoder.
PixArt Models
- PixArt-XL-2-1024-MS: Best for high-resolution images (1024px).
- PixArt-Sigma-XL-2-1024-MS: Optimized for 1024px with SDXL VAE.
- PixArt-LCM-XL-2-1024-MS: Uses LCM model for specific artistic styles.
HunYuan DiT Models
- HunYuanDiT.pt: Main model file for HunYuan DiT.
- mT5-xl.bin: Text encoder for HunYuan DiT.
- chinese-roberta-wwm-ext-large.bin: Another text encoder for HunYuan DiT.
DiT Models
- DiT-XL-2-256x256: For 256x256 images.
- DiT-XL-2-512x512: For 512x512 images.
T5 Models
- t5-v1_1-xxl: Main T5 model for text encoding.
MiaoBi Models
- MiaoBi_CLIP.safetensors: CLIP model for MiaoBi.
- MiaoBi.safetensors: UNet model for MiaoBi.
VAE Models
- vae-ft-mse-840000-ema-pruned.safetensors: For SD1.X, DiT, and PixArt alpha.
- sdxl-vae-fp16-fix.safetensors: For SDXL and PixArt sigma.
Common Issues and Solutions
- Model Not Loading:
- Ensure the model files are placed in the correct directory.
- Verify that the model file names match the expected names in the documentation.
- Poor Image Quality:
- Install xformers for better performance.
- Adjust the CFG and step settings as recommended for each model.
- High Memory Usage:
- Use FP16 or bnb4bit mode to reduce VRAM usage.
- Offload models to CPU if you have limited GPU memory.
Frequently Asked Questions
- Q: Can I use multiple models simultaneously?
- A: Yes, you can switch between models depending on your needs, but only one model can be active at a time.
- Q: How do I update the models?
- A: Use the
git pull
command in the custom nodes directory to update the models.
For more detailed tutorials, documentation, and community support, visit the following resources:
-
These resources provide comprehensive guides, sample workflows, and community forums where you can ask questions and share your experiences.