Instructions to use lightx2v/LightVAE with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Wan2.2
How to use lightx2v/LightVAE with Wan2.2:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
ComfyUI
let’s pretend I didn’t read the model card :)
https://huggingface.co/rzgar/r_drive/tree/main/LightVAE
lightvae-lite-h3-encoder.safetensors can not loaded by 'load vae'node , but the kvs all the same
elif "decoder.transformer_blocks.0.scale1" in sd and "encoder.down.5.block.0.conv1.weight" in sd: # MiniMax H3 video VAE
https://huggingface.co/lightx2v/LightVAE/blob/main/lightvae-lite-h3-decoder-comfy.safetensors
# ComfyUI Error Report
## Error Details
- **Node ID:** 7212
- **Node Type:** VAELoader
- **Exception Type:** RuntimeError
- **Exception Message:** RuntimeError: ERROR: VAE is invalid: None
If the VAE is from a checkpoint loader node your checkpoint does not contain a valid VAE.
Thanks for the feedback, and sorry for the confusing installation instructions. We’ve simplified the H3 integration and clarified the setup.
The repository root is not a ComfyUI node, which explains the reported missing __init__.py error. Clone it outside custom_nodes, then run the installer using ComfyUI’s Python environment:
cd /path/to/ComfyUI
git clone https://github.com/ModelTC/LightVAE.git LightVAE
python LightVAE/comfyui/install.py --comfy-root .
For an existing checkout, run git -C LightVAE pull and rerun the installer.
This creates custom_nodes/ComfyUI-LightVAE/. Restart ComfyUI and search for Load H3 LightVAE. No LightVAE package installation or custom Triton kernels are needed.
The Lite files are separate components, not complete VAEs. Load the original H3 video VAE with the standard Load VAE node and connect it to base_vae. Select the Lite components in Load H3 LightVAE instead.
- FL2V: replace only the decoder; keep original keyframe encoding.
- REF2V: replace the encoder, decoder, or both independently. Encoder-only replacement works with
decoder = none. - Keep the audio VAE unchanged.
Both the standard H3 checkpoints and previous H3 -comfy exports are supported.
Installation, downloads and example workflows: H3 ComfyUI guide.
I thought we can use the encoder seperately. maybe you can make a comparision with minimax_h3_lynnreal_light_vae_int8_convrot, which i think smaller but a little faster
Our
-comfy.safetensorsfiles require the Load LightVAE (TLR) custom node—not ComfyUI’s standard Load VAE / VAELoader.
You mean this:
https://github.com/ModelTC/LightVAE ?
Cannot be loaded because the init py file is missing.
FileNotFoundError: [Errno 2] No such file or directory: 'D:\ComfyUI\custom_nodes\LightVAE\init.py'
@GlamoramaAttack OMG it's because the node is in the comfyui folder
@lightx2v could you have made this more of a pain? Can't just do a simple checkout, if you're going to put a out a node, at least do it how everyone who uses ComfyUI expects please
yes , this is rather complicated for a vae change ...