edit: 1.1 can work with long vídeos, long context, mask with sam, this is veeeeeeeeeeeery best than 1.0 try xD!

MiniMax H3 Character Replacement — Ref2VA + Turbo

A simple ComfyUI workflow for full character replacement using MiniMax H3 Ref2VA.

Provide a source video and a reference image of the new character. The workflow uses the original video as reference for motion, poses, camera movement, timing and scene composition, while replacing the original character with the identity and appearance from the reference image.

⚡ Uses the MiniMax H3 Turbo LoRA (Pruned) for fast 4-step generation, making character replacement significantly faster than the standard 20-step workflow.

Inputs:

• 🎬 Source Video — motion, camera and scene reference

• 🖼️ Character Image — replacement character

• ✍️ Prompt — optional adjustments

Model: MiniMax H3 Ref2VA Pruned INT8 ConvRot
Sampler: MiniMax H3 Turbo
Steps: 4