MiniMax H3 character swap setup in ComfyUI

This tutorial explains how to set up a MiniMax H3 character replacement workflow in ComfyUI. The presenter describes a local AI setup that needs over 40 GB of model files and a strong GPU. Installation covers importing the workflow JSON, installing missing nodes, downloading models and restarting ComfyUI to clear remaining errors.

The workflow takes a source video and a replacement character image. Viewers learn to set frame limits, skip opening frames and match the output aspect ratio. The presenter states that H3 supports clips up to 15 seconds and that the workflow converts input to 24 fps. One background mode keeps the video's scene; the other uses the image's scene while following the video's movement and camera framing.

Prompt generation uses either ComfyUI's paid Gemini node, which requires an account and credits, or a manual chatbot route demonstrated with ChatGPT. The latter involves uploading the video and image, then copying the generated prompt into ComfyUI. Normal mode uses 20 steps; Turbo provides a faster preview option.

The presenter reports roughly 10 minutes to generate a five-second clip on his machine, without specifying its hardware. His examples cover colored lighting, foreground objects and cuts. He notes that lip sync works best in close-ups and that distant characters can look pixelated.