ComfyUI-H3VAE_TRT

Running TensorRT version of the MiniMax-H3 VAE in ComfyUI, which can increase speed by up to 1.7x

cd ComfyUI/custom_nodes
git clone https://github.com/lihaoyun6/ComfyUI-H3VAE_TRT.git
python -m pip install -r ComfyUI-H3VAE_TRT/requirements.txt

1. Download encoder & decoder .onnx files (and .data files, if present) -> here

   You can use the w4a16_awq decoder if you don't have 12GB+ of VRAM.

2. Put models into ComfyUI/models/vae

• Compile the TensorRT engine from onnx using the MiniMax-H3 TRT VAE Compiler node before first use.
• After compiling the TRT Engines, you can use the MiniMax-H3 TRT VAE Loader node to load them.

• ComfyUI @comfyanonymous
• MiniMax-H3 @MiniMax-AI