ComfyUI-DreamFit

ComfyUI-DreamFit

Garment-centric human generation nodes for ComfyUI using DreamFit with Flux.

DreamFit is a powerful adapter system that enhances Flux models with garment-aware generation capabilities, enabling high-quality fashion and clothing generation.

Features

• 🎨 Garment-Centric Generation: Generate humans wearing specific garments with high fidelity
• 👗 Virtual Try-On: Try different garments on models
• 🎭 Pose Control: Generate with specific poses while maintaining garment details
• 🔧 Flux Integration: Seamlessly works with Flux models in ComfyUI
• ⚡ Adaptive Attention: Smart injection of garment features into the generation process
• 🎯 LoRA Adaptation: Efficient model adaptation without full fine-tuning

Installation

Prerequisites

• ComfyUI installed and working
• Python 3.8 or higher
• Git (for installation method 1)

Method 1: Git Clone (Recommended)

# Navigate to your ComfyUI custom nodes directory
cd ComfyUI/custom_nodes

# Clone the repository
git clone https://github.com/skayka/ComfyUI-DreamFit.git

# Navigate to the installed directory
cd ComfyUI-DreamFit

# Install Python dependencies
pip install -r requirements.txt

# Optional: Download models (do this after restarting ComfyUI)
python download_models.py

Method 2: ComfyUI Manager

If you have ComfyUI Manager installed:

1. Open ComfyUI Manager
2. Search for "DreamFit"
3. Click Install
4. Restart ComfyUI

Method 3: Manual Installation

1. Download the repository as ZIP
2. Extract to ComfyUI/custom_nodes/ComfyUI-DreamFit
3. Open a terminal in the extracted directory
4. Install requirements:

   pip install -r requirements.txt
   

Download Models

After installation, download the required DreamFit models:

# Navigate to the ComfyUI-DreamFit directory
cd ComfyUI/custom_nodes/ComfyUI-DreamFit

# Download all models (855MB total)
python download_models.py

# Or download specific models
python download_models.py --model flux_i2i
python download_models.py --model flux_tryon

# List available models
python download_models.py --list

# Verify downloaded models
python download_models.py --verify

The models will be automatically downloaded to:

• If inside ComfyUI: ComfyUI/models/dreamfit/
• Otherwise: ./dreamfit_models/

Troubleshooting Installation

"No module named 'folder_paths'" during install

This is normal! The installation will complete successfully. This error only appears because the installer tries to verify the installation outside of ComfyUI.

Models not found

1. Make sure you've downloaded the models using python download_models.py
2. Check that models are in ComfyUI/models/dreamfit/
3. Restart ComfyUI after downloading models

Import errors in ComfyUI

1. Make sure you've installed requirements: pip install -r requirements.txt
2. Restart ComfyUI after installation
3. Check the ComfyUI console for specific error messages

Available Nodes

1. DreamFit Checkpoint Loader

Loads DreamFit model checkpoints and initializes the Anything-Dressing Encoder.

Inputs:

• model_name: Choose from available DreamFit models
• device: Processing device (cuda/cpu)
• dtype: Model precision (fp16/bf16/fp32)

Outputs:

• DREAMFIT_MODEL: The loaded model configuration
• DREAMFIT_ENCODER: The Anything-Dressing Encoder
• DREAMFIT_CONFIG: Model configuration

2. DreamFit Encode

Encodes garment images and text prompts into conditioning for generation.

Inputs:

• encoder: From checkpoint loader
• garment_image: The garment to generate
• positive_prompt: Description of desired output
• negative_prompt: What to avoid in generation
• model_image (optional): Reference pose/model
• garment_description: Brief garment description
• garment_category: Type of garment (casual/formal/sportswear/traditional)
• enhance_prompt: Auto-enhance prompts
• use_model_parse: Parse model image for better results
• injection_strength: Control garment influence (0.1-2.0)

Outputs:

• conditioning: DreamFit conditioning for sampling
• enhanced_prompt: Improved positive prompt
• enhanced_negative: Improved negative prompt

3. DreamFit Flux Adapter V2

Applies DreamFit adaptation to a Flux model with CLIP integration.

Inputs:

• model: Your Flux model
• clip: CLIP model for text encoding
• dreamfit_conditioning: From encode node
• positive: Enhanced positive prompt
• negative: Enhanced negative prompt
• lora_strength: LoRA adaptation strength (0.0-2.0)
• injection_strength: Feature injection strength (0.0-2.0)
• lora_merge_mode: How to merge LoRA weights
• injection_mode: Attention injection strategy
• attention_mode: Which attention layers to modify
• use_cached_embeddings: Speed optimization

Outputs:

• model: Flux model with DreamFit adaptation
• positive: Positive conditioning
• negative: Negative conditioning

4. DreamFit K-Sampler

Custom sampler optimized for DreamFit generation.

Inputs:

• Standard KSampler inputs (model, seed, steps, cfg, etc.)
• dreamfit_conditioning: From encode node

Outputs:

• LATENT: Generated image latent

5. DreamFit Unified

Complete DreamFit integration in a single node.

Inputs:

• model: Flux diffusion model (from UNETLoader)
• positive/negative: Pre-encoded conditioning from CLIP
• garment_image: Garment to process
• dreamfit_model: Select model type
• strength: Overall adaptation strength
• model_image (optional): Reference pose for try-on
• injection_strength: Garment feature strength
• injection_mode: Feature injection strategy

Outputs:

• model: Enhanced Flux model
• positive/negative: Enhanced conditioning
• debug_garment: Processed garment (224x224) for debugging

6. DreamFit Simple

All-in-one node for easy DreamFit generation.

Inputs:

• model: Flux model
• clip: CLIP model
• vae: VAE model
• dreamfit_model: Select DreamFit model
• garment_image: Garment to generate
• positive/negative: Text prompts
• seed, steps, cfg, denoise: Standard generation parameters
• model_image (optional): Reference model/pose

Outputs:

• samples: Generated latent image

6. DreamFit Sampler Advanced

Advanced sampler with additional controls.

Features:

• Noise modes: default, garment_aware, structured
• Injection schedules: constant, linear, cosine, step
• Step control for multi-stage generation

Workflow Examples

Four example workflows are included in the workflows/ directory:

1. Simple Workflow (dreamfit_simple_workflow.json)

• Uses the all-in-one DreamFit Simple node
• Minimal setup required
• Best for quick testing and basic generation

2. Basic Workflow (dreamfit_basic_workflow.json)

• Full node setup with individual components
• More control over the generation process
• Good balance of simplicity and flexibility

3. Advanced Workflow (dreamfit_advanced_workflow.json)

• Complete setup with all advanced features
• Virtual try-on with model images
• Advanced sampling strategies
• Maximum control and customization

Loading Workflows

1. Open ComfyUI
2. Click "Load" in the menu
3. Navigate to ComfyUI/custom_nodes/ComfyUI-DreamFit/workflows/
4. Select the desired workflow JSON file

4. Unified Workflow (dreamfit_unified_workflow.json)

• Uses the new DreamFit Unified node
• Proper Flux model loading (UNETLoader, DualCLIPLoader, VAELoader)
• Shows debug output of processed garment
• Best for understanding the complete pipeline

Tips for Best Results

1. Image Quality: Use high-quality garment images with clean backgrounds
2. Prompts: Be specific about the desired style and context
3. Injection Strength: Start with 0.5 and adjust based on results
4. Sampling Steps: 20-30 steps usually sufficient
5. CFG Scale: 7-8 works well for most cases

Model Information

| Model | Size | Description | Best For | |-------|------|-------------|----------| | flux_i2i | 284MB | Basic garment generation | General fashion images | | flux_i2i_with_pose | 284MB | Pose-controlled generation | Specific poses/positions | | flux_tryon | 287MB | Virtual try-on | Trying clothes on models |

Troubleshooting

"No module named 'dreamfit_core'"

• Ensure you're in the correct directory when installing
• Try python -m pip install -e . from the ComfyUI-DreamFit directory

Models not downloading

• Check internet connection
• Try running with specific model: python download_models.py --model flux_i2i
• Manually download from HuggingFace and place in ComfyUI/models/dreamfit/

Out of memory errors

• Reduce batch size to 1
• Use CPU offloading in ComfyUI settings
• Try the fp16 versions of models

Technical Details

DreamFit uses:

• Anything-Dressing Encoder: 83.4M parameter encoder for garment features
• Adaptive Attention Injection: Injects garment features into Flux attention layers
• LoRA Adaptation: Efficient 16-rank LoRA for model adaptation

Credits

• Original DreamFit: ByteDance Research
• ComfyUI: comfyanonymous

License

This project follows the same license as the original DreamFit implementation.

Contributing

Contributions are welcome! Please:

1. Fork the repository
2. Create a feature branch
3. Submit a pull request

Support

• Report issues on GitHub Issues
• Join the discussion in ComfyUI Discord