2025-08-03 15:26:22 +02:00
2025-08-03 15:26:22 +02:00
2025-08-03 02:12:40 +02:00
2025-08-03 15:26:22 +02:00
2025-08-03 02:14:33 +02:00

WanImageToVideoTiledVAE for ComfyUI

This is a custom node for ComfyUI that replaces the WanImageToVideo node but uses a Tiled VAE approach to reduce VRAM requirements.

Description

WanImageToVideoTiledVAE is designed to be a drop-in replacement for the WanImageToVideo node in ComfyUI. By using a tiled VAE, it significantly reduces the VRAM needed, making it more accessible for users with limited resources. I found it very useful for ComfyUI-zluda, since the VAE is particularly VRAM-hungry there.

Installation

To install this node, follow these steps:

  1. Clone this repository into your ComfyUI custom nodes directory.
  2. Restart ComfyUI to load the new node.
git clone https://github.com/stduhpf/ComfyUI--WanImageToVideoTiled.git /path/to/ComfyUI/custom_nodes/WanImageToVideoTiledVAE

Usage

To use this node in ComfyUI:

  1. Add the WanImageToVideoTiledVAE node to your workflow.
  2. Connect the required inputs (positive and negative conditioning, VAE model, start image).
  3. Optionally, connect a start image or clip vision output.
  4. Run the workflow.

Inputs

Input Type Description
positive CONDITIONING Positive conditioning input.
negative CONDITIONING Negative conditioning input.
vae VAE The VAE model to use.
width INT Width of the video, default is 832.
height INT Height of the video, default is 480.
length INT Number of video frames, default is 81.
batch_size INT Batch size, default is 1.
tile_size INT Size of the tiles, default is 512.
overlap INT Overlap between tiles, default is 64.
temporal_size INT Amount of frames to encode at a time, default is 64.
temporal_overlap INT Amount of frames to overlap, default is 8.
start_image IMAGE (optional) Optional start image.
clip_vision_output CLIP_VISION_OUTPUT (optional) Optional clip vision output.

Outputs

Output Type Description
latent LATENT Empty latents with start_image encoded as first frame.
positive CONDITIONING Processed positive conditioning.
negative CONDITIONING Processed negative conditioning.

License

This project mostly contains code copy-pasted from ComfyUI, which is licenced under GPL3.0. Therefore it is also licenced under GPL 3.0. (see LICENCE file for more details)

S
Description
No description provided
Readme GPL-3.0
82 KiB
Languages
Python 100%