ReActor | Fast Face Swap

1. What is the ComfyUI ReActor Face Swap Workflow?

The ComfyUI ReActor Face Swap workflow enables seamless face swapping in images and videos using advanced face detection, swapping, and enhancement models. This ReActor Face Swap implementation produces natural, high-quality results while providing flexible restoration and upscaling options.

Latest Updates

Recent updates have brought significant enhancements to ReActor Face Swap workflow. Support for GPEN 1024/2048 restoration models improves face quality, while updates to video processing and overall performance optimization enable faster face-swapping operations.

2. Benefits of ComfyUI ReActor Face Swap:

• Produces realistic face replacements while maintaining natural facial expressions, lighting, and texture
• Offers flexible restoration features with customizable enhancement settings
• Includes upscaling support to improve image and video resolution for clearer outputs
• Delivers smooth video outputs using frame interpolation for fluid motion

3. How to Use the ComfyUI ReActor Face Swap Workflow

3.1 Generation Methods with ComfyUI ReActor Face Swap

Primary Method: Video-to-Image Face Replacement

• Inputs:
  • Target Video: Loaded via the Load Video (Upload) node
  • Source Image: Loaded via the Load Image node
• Procedure:
  • Run the ReActor Face Swap workflow by clicking Queue Prompt for high-quality results
  • The final video will be saved using the Video Combine node

Alternative Methods

The ReActor Face Swap workflow supports both multi-face and image-only processing. For multiple face swap, adjust the input_faces_index parameter in the Fast Face Swap node to target specific faces (e.g., 0,1,2). For image-only operations, simply use the Load Image node instead of Load Video (Upload) for your target input.

4. Nodes and Parameters Reference of ComfyUI ReActor Face Swap

Load Video (Upload) Node

• Purpose: Loads the target video for face swapping
• Key Parameters:
  • force_rate: Adjusts the frame rate for video extraction
  • select_every_nth: Controls frame selection frequency
    • Lower values: More frames, smoother but longer processing
    • Higher values: Fewer frames, faster but choppier
  • skip_first_frames: Skips specified number of initial frames
    • Useful for trimming input video start
• Trade-offs: Balance between video smoothness and processing time

Load Image Node

• Purpose: Loads the source image containing the face to swap
• Key Parameters:
  • image: Path to the source face image
    • Ensure high-quality, clear face images for best results
• Trade-offs: Image quality directly impacts swap results

Fast Face Swap Node

• Purpose: Performs the face swapping operation
• Key Parameters:
  • swap_model: Face swapping model selection
    • inswapper_128.onnx: Default model optimized for quality
  • facedetection: Face detection algorithm choice
    • retinaface_resnet50: High accuracy, slower
    • retinaface_mobile0.25: Faster, lower precision
  • face_restore_visibility: Restoration blend strength (0.1–1.0)
    • Higher: Stronger restoration, potential over-smoothing
    • Lower: Subtle restoration, preserves original details
  • codeformer_weight: Detail preservation balance (0.0–1.0)
    • Higher: Smoother results, less detail
    • Lower: More details, potential artifacts
  • input_faces_index: Specifies faces to swap in multi-face processing
• Trade-offs: Balance between quality, speed, and accuracy

Face Booster Node

• Purpose: Enhances and restores swapped face before integration
• Key Parameters:
  • interpolation: Enhancement method selection (e.g., Bicubic)
  • visibility: Restoration intensity (0.1–1.0)
    • Higher: Stronger enhancement but may over-process
    • Lower: Subtle enhancement, more natural look
  • codeformer_weight: Detail balance (0.0–1.0)
    • Higher: Smoother, less detail
    • Lower: More detail, possible artifacts
• Trade-offs: Enhanced quality vs natural appearance

Load Upscale Model Node

• Purpose: Loads AI upscaling model for resolution enhancement
• Key Parameters:
  • model_name: Upscaling model selection (e.g., RealESRGAN_x4plus.pth)
• Trade-offs: Higher resolution vs memory usage

Upscale Image Node

• Purpose: Applies upscaling to processed images
• Key Parameters:
  • upscale_model: Selected model application
  • image: Target for upscaling
• Trade-offs: Quality improvement vs processing time

Video Combine Node

• Purpose: Combines processed frames into final video
• Key Parameters:
  • frame_rate: Output video frame rate
    • Higher: Smoother playback, more frames needed
    • Lower: Choppier but faster processing
  • filename_prefix: Custom prefix for output file
  • save_output: Controls video saving to specified path
• Trade-offs: Output quality vs file size

5. More Information

For updates, model details, and ReActor Face Swap workflow support, please visit the official GitHub Repository https://github.com/Gourieff/comfyui-reactor-node.