WanVaceToVideoMC - Multi-Control Enhancement for ComfyUI

A memory-efficient enhancement of the WanVaceToVideo node that supports multiple control inputs (pose, depth, edge) in a single node, solving the VAE triple-loading memory crisis.

Features

• Single VAE Instance: Uses one VAE for all control processing, saving ~20-30GB VRAM
• Multi-Control Support: Process pose, depth, and edge controls simultaneously
• Granular Control: Independent strength controls for video and mask per control type
• Full Backward Compatibility: Works seamlessly with existing WanVaceToVideo workflows
• Multiple Combination Modes: Choose how controls are combined (multiply, add, average, max)

Installation

1. Navigate to your ComfyUI custom nodes directory:

   cd ComfyUI/custom_nodes/

2. Clone this repository:

   git clone https://github.com/yourusername/WanVaceToVideoMC.git

3. Restart ComfyUI

Usage

Multi-Control Mode

The node appears as "WAN VACE to Video (Multi-Control)" in the node menu under conditioning/video_models.

New inputs for multi-control:

• control_video_pose / control_masks_pose - Pose control inputs
• control_video_depth / control_masks_depth - Depth control inputs
• control_video_edge / control_masks_edge - Edge control inputs
• strength_video_* - Control strength for video influence (0.0-10.0)
• strength_mask_* - Control strength for mask influence (0.0-10.0)
• multi_control_mode - How to combine controls: multiply, add, average, max

Legacy Mode

For backward compatibility, you can still use the original inputs:

• control_video - Single control video input
• control_masks - Single control mask
• reference_image - Reference image for style

Note: You cannot use both legacy and multi-control inputs simultaneously.

Memory Savings

Traditional approach (3 separate WanVaceToVideo nodes):

• VAE memory usage: ~30-45GB (3 × 10-15GB)

WanVaceToVideoMultiControl:

• VAE memory usage: ~10-15GB (single instance)
• Savings: ~20-30GB VRAM

Control Combination Modes

• multiply: Most restrictive - all controls must agree
• add: Most permissive - any control can influence
• average: Balanced combination (default)
• max: Strongest signal wins

Requirements

• ComfyUI (latest version)
• PyTorch >= 2.0.0
• CUDA-capable GPU (tested on 3x A6000 setup)

Credits

Developed at Zerospace for production workflows with WAN VACE 14B video generation.

License

AGPL-3.0 License - See LICENSE file for details.