Extensions/ComfyUI-H3VAE_TRT
ComfyUI Extension

ComfyUI-H3VAE_TRT

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

By lihaoyun6·Created 21 days ago·Updated 16 days ago· 242
lihaoyun6/ComfyUI-H3VAE_TRT
Nodes2
On cloudLocal install
CategoryMiniMax_H3/Acceleration
Stars242
Updated16 days ago

Nodes (2)

MiniMax-H3 TRT VAE Compiler

Compile the MiniMax-H3 VAE to TensorRT — the one-time job that pays for itself

MiniMax_H3/Acceleration
MiniMax-H3 TRT VAE Loader

The TensorRT MiniMax-H3 VAE loader — how your H3 videos get fast without looking like it

MiniMax_H3/Acceleration
Readme

ComfyUI-H3VAE_TRT

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

Preview

Installation

Install the node:

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

Usage

Download Models

  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

Nodes

  • 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.

    If you need to compile outside of ComfyUI, use the compile.py script in the project.

Credits