🎬 Professional Video-to-Video Transformation with WAN VACE

Transform your videos with professional quality using this comprehensive ComfyUI workflow for WAN VACE. This complete pipeline enables seamless video-to-video transformation of long-form videos with advanced features including seamless joining, upscaling, and frame interpolation. Break down lengthy videos into manageable segments, process them individually, and seamlessly combine them back into cohesive, high-quality output.

✨ Key Features

• Long Video Processing: Handle extended video content by breaking into segments and seamlessly rejoining

• Complete V2V Pipeline: Full video-to-video transformation workflow

• Seamless Video Joining: Custom nodes for professional video concatenation without visible transitions

• Multi-Step Process: Generate → Join → Combine → Upscale → Interpolate

• Professional Quality: High-quality output with customizable settings

• Memory Optimization: Low VRAM options for various GPU configurations

• Batch Processing: Process multiple video segments efficiently

• Scalable Architecture: Handle videos of any length through intelligent segmentation

📋 Requirements

Essential Model Files

🔴 WAN GGUF Models

• Download from: QuantStack/Wan2.1_T2V_14B_FusionX_VACE-GGUF

• Choose your preferred quantization (Q3_K_S, Q8_0, etc.)

• Place in: ComfyUI/models/unet

🟣 WAN VAE

• Download wan_2.1_vae.safetensors from: Comfy-Org/Wan_2.1_ComfyUI_repackaged

• Place in: ComfyUI/models/vae

🟣 WAN Text Encoder

• Download GGUF text encoders from: city96/umt5-xxl-encoder-gguf

• Place in: ComfyUI/models/text_encoders

Required Custom Nodes

⚠️ Important: Download these custom nodes from this page (not available in ComfyUI-Manager):

• seamless_join_video_clips.py

• combine_video_clips.py

• Place in: ComfyUI/custom_nodes/

ComfyUI Extensions

⚙️Install these custom notes using the ComfyUI-Manager.

• ComfyUI-GGUF

• ComfyUI-VideoHelperSuite

• ComfyUI-KJNodes

• ComfyUI-ControlNet-Aux

• ComfyUI-Frame-Interpolation

• ComfyUI-Easy-Use

📖 Step-by-Step Guide

Initial Setup

1. Configure Constants:

   • Width/Height: 576x1024 (9:16 aspect ratio) or match your source video

   • Length: 81 frames per segment

   • Skip Frames: Start with 0

   • Filename Prefix: Set your output folder and prefix

2. Load Source Materials:

   • Load your source video for restyling

   • Load reference image (ensure similar pose to first video frame)

   • Use SDXL/FLUX with LoRA and ControlNet for best pose matching

Step 1: Generate WAN Videos

1. Write Prompts:

   • Describe subject, outfit, and background

   • Include action phrases for dynamic results

2. Generate Video Segments:

   • Click run to generate first 81-frame video segment

   • Increase skip frames by 81 to process next segment

   • Repeat for the entire length of your source video

   • Final segment can be shorter but may have lower quality

   • For long videos: Continue this process until you've covered the full duration

Step 2: Join Videos Seamlessly

1. Configure Joining:

   • Set folder path to your generated videos

   • Set filename prefix matching your generated files

   • Start with filename suffix = 1

   • Use same prompt from Step 2

2. Join Process:

   • Run to join first and second videos

   • Increase filename suffix by 1

   • Run to join second and third videos

   • Repeat until all segments are joined

Step 3: Combine, Upscale, and Interpolate

1. Final Processing Setup:

   • Set folder path to joined videos

   • Keep filename suffix = 1 (constant)

   • Set combine filename for final output

   • Set upscale filename for enhanced version

2. Execute Final Pipeline:

   • Combine all joined videos

   • Upscale using RealESRGAN (2x scale)

   • Interpolate frames using FILM VFI (2x frame rate)

⚙️ Advanced Settings

Low VRAM Configuration

• Use the UnetLoaderGGUFDisTorchMultiGPU node for memory optimization

• Set virtual_vram_gb to 2.0-4.0 for 12GB and lower GPUs

• Enable use_other_vram for additional memory fallback

Performance Optimization

• Bypass PathchSageAttentionKJ and ModelPatchTorchSettings if you don't have Triton

• Adjust batch sizes based on your GPU memory

• Use appropriate quantization levels for your hardware

🎯 Tips for Best Results

1. Long Video Strategy: Plan your segmentation approach - 81 frames per segment ensures smooth transitions while maintaining manageable processing chunks

2. Reference Image Quality: Use high-quality reference images with poses similar to your source video's first frame

3. Prompt Engineering: Be specific about subject details, clothing, and background elements

4. Segment Planning: Plan your video segments to maintain narrative continuity across the entire video length

5. Hardware Considerations: Adjust settings based on your GPU capabilities - longer videos benefit from optimized VRAM settings

6. Consistency Maintenance: Keep prompts consistent across all segments to ensure visual coherence in the final long video

🩺 Troubleshooting

• OOM Errors: Increase virtual_vram_gb or reduce batch sizes

• Missing Nodes: Ensure all custom nodes are properly installed

• Quality Issues: Check reference image alignment and prompt specificity

• Processing Slow: Consider using lower quantization models for faster generation

🔧 Custom Nodes Parameter Guide

WanVideoVaceSeamlessJoin Node

This custom node seamlessly joins two video clips with intelligent masking for smooth transitions.

Parameters:

• mask_last_frames (INT): Number of frames to mask at the end of the first video

  • Default: 0

  • Range: 0-20

  • Use 0 for no masking, 5-10 for subtle blending

• mask_first_frames (INT): Number of frames to mask at the beginning of the second video

  • Default: 10

  • Range: 0-20

  • Recommended: 10 frames for smooth transitions

• frame_load_cap (INT): Maximum number of frames to load from each video

  • Default: 81

  • Range: 1-1000

  • Should match your segment length (typically 81)

• first_video_path (STRING): Full path to the first video file

  • Format: "C:\path\to\video1.mp4"

  • Use absolute paths for reliability

• second_video_path (STRING): Full path to the second video file

  • Format: "C:\path\to\video2.mp4"

  • Ensure file exists and is accessible

Outputs:

• image: Combined video frames as image sequence

• mask: Generated mask for the transition area

CombineVideoClips Node

This node combines multiple video clips into a single continuous sequence with advanced masking options.

Parameters:

• frame_load_cap (INT): Maximum frames to load per video

  • Default: 81

  • Range: 1-1000

  • Should match your segment frame count

• mask_last_frames (INT): Frames to mask at the end of each video (except last)

  • Default: 0

  • Range: 0-20

  • Use 0 for clean cuts, 5-10 for fade effects

• mask_first_frames (INT): Frames to mask at the beginning of each video (except first)

  • Default: 10

  • Range: 0-20

  • Recommended: 10 for smooth transitions

• first_video_path (STRING): Path to the first video in sequence

  • Base video - typically your original generated video

• first_joined_video_path (STRING): Path to first seamlessly joined video

  • Result from first WanVideoVaceSeamlessJoin operation

• second_joined_video_path (STRING): Path to second seamlessly joined video

  • Result from second WanVideoVaceSeamlessJoin operation

• third_joined_video_path (STRING): Path to third seamlessly joined video

  • Continue pattern for additional segments

• fourth_joined_video_path (STRING): Path to fourth seamlessly joined video

  • Optional - use if you have this many segments

• fifth_joined_video_path (STRING): Path to fifth seamlessly joined video

  • Optional - maximum supported segments

• last_video_path (STRING): Path to the final video in sequence

  • The last generated video segment

Output:

• image: Combined video sequence as image frames ready for final processing

Parameter Optimization Tips:

For Seamless Joining:

• Short transitions: mask_first_frames = 5, mask_last_frames = 0

• Smooth blending: mask_first_frames = 10, mask_last_frames = 5

• Long crossfades: mask_first_frames = 15, mask_last_frames = 10

For File Paths:

• Ensure all video files exist before running

• Use consistent naming conventions for easier batch processing

Frame Count Considerations:

• Set frame_load_cap to match your segment length (usually 81)

• Smaller values may truncate longer segments

This workflow provides professional-grade video transformation capabilities with comprehensive control over the entire pipeline from generation to final output.