# MiniMax H3 in ComfyUI | Character Swap, Talk, Sing & Timeline (Ep41) — retained record

> Retained reddit record from reddit:comfyui.

Canonical: https://highsignal.app/data/records/7f6d373625d793a8

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:comfyui)  published 2026-11-03 16:07 ingested 2026-11-04 02:41

[original source ↗](https://www.reddit.com/r/comfyui/comments/1xmx04k/minimax_h3_in_comfyui_character_swap_talk_sing/)

## retained text



```
In this episode, we test MiniMax H3 in ComfyUI with 25 free workflows, plus 2 bonus workflows. H3 is a video model that makes video with sound, from your images, videos and audio. We test them all, and see what H3 does well and where it struggles.



Part 1, change your video: a character sheet from one picture, then swap the person in your video (a character, only the head, or a cartoon and mascot), motion transfer, edit a video by text, change the outfit from pictures, restyle a video, sharpen it to Full HD, and join your clips into one video with their sound.



Part 2, bring a picture to life: tell it what to do, make a picture talk from typed words, make it talk or sing with your own voice or song (Audio Sync), and an outfit change spin with Flux Klein and H3.



Part 3, long videos with H3 Timeline: sing a song in scenes, clean cuts, four singers in one song, a character who walks in, act a story, walk into a new place, redo one scene and keep the good ones, and a bigger size.



Part 4, remake your video in 2K, plus two bonus workflows: text to video in 2K, and first and last frame to video.



We also look honestly at where H3 struggles: a person who joins a scene can come in twice, one seed can change the whole look of a scene, and Talk keeps the pose of your picture.
```



document fetched 2026-11-04 00:41 · canonical https://reddit.com/r/comfyui/comments/1xmx04k/minimax_h3_in_comfyui_character_swap_talk_sing
