ComfyUI-VAE-Utils

KO | EN

gitlite — search

#python #hacktoberfest #react #golang #android #deep-learning #docker #nodejs #awesome #kubernetes #pytorch #openai #mcp #ios

(★ 224)

No description available.

ComfyUI-VAE-Utils 최신버젼 다운로드

// repository documentation

★ 0 (0 ratings)

![side by side comparison](https://raw.githubusercontent.com/spacepxl/ComfyUI-VAE-Utils/main/assets/cover.png)

Nodes for loading and using VAEs in ways not supported by base ComfyUI. Currently supports Wan upscale VAE, more may be added later as needed.

## Install

```
cd custom_nodes
git clone https://github.com/spacepxl/ComfyUI-VAE-Utils
```

That's it! No extra requirements outside of what ComfyUI already uses.

## Models

[Wan2.1 VAE 2x imageonly real v1](https://huggingface.co/spacepxl/Wan2.1-VAE-upscale2x/blob/main/Wan2.1_VAE_upscale2x_imageonly_real_v1.safetensors)

## Nodes

### Load VAE (VAE Utils)

Direct replacement for native Load VAE node, but with added input/output channel detection for Wan2.1 VAE

### VAE Decode (VAE Utils)

Decode latents with optional auto detected upscaling, and optional tiling

### Latent Upscale (VAE Utils)

![latent upscale comparison](https://raw.githubusercontent.com/spacepxl/ComfyUI-VAE-Utils/main/assets/wan21_latent_upscale_2x.jpg)

Simple neural latent upscaler (right), far from optimal but much better than naive bilinear/bislerp interpolation (left).

Currently supported: Wan2.1 2x

## Workflows

[workflow/workflow_wan_t2i_upscale2x.png](https://github.com/spacepxl/ComfyUI-VAE-Utils/blob/main/workflow/workflow_wan_t2i_upscale2x.png)

## IMPORTANT

To use native VAE decoding with latents from kijai/ComfyUI-WanVideoWrapper, you need to un-normalize the latents first using this node:

![WanVideoWrapper latent rescale](https://raw.githubusercontent.com/spacepxl/ComfyUI-VAE-Utils/main/assets/WanVideoWrapper_latent_rescale.png)