🎬 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

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
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

Write Prompts:
- Describe subject, outfit, and background
- Include action phrases for dynamic results
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

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
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

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
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

Long Video Strategy: Plan your segmentation approach - 81 frames per segment ensures smooth transitions while maintaining manageable processing chunks
Reference Image Quality: Use high-quality reference images with poses similar to your source video's first frame
Prompt Engineering: Be specific about subject details, clothing, and background elements
Segment Planning: Plan your video segments to maintain narrative continuity across the entire video length
Hardware Considerations: Adjust settings based on your GPU capabilities - longer videos benefit from optimized VRAM settings
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.

Search AI Models & Addons

very few results