← sources

V5 Update Newbie Minimax H3 sam 3.1 latent mask for guided character replacement

source: reddit (reddit:StableDiffusion)published 2026-09-21 07:39ingested 2026-09-22 02:38

retained text

the first video is a side by side of original video - char replacement using sam 3.1 - char replacement without sam 3.1.

if you look you'll see that  both 1 and 2 are in sync in position and coloring of background, the 3rd is slightly off color, off position to show how using a latent mask can affect char replacement.

[https://github.com/roycho87/minimax\_wf/tree/main](https://github.com/roycho87/minimax_wf/tree/main)

* I added a block audio feature that will force minimax to change your voice everytime when editing video.
* Resolution selector with 2 decimal spaces so you can actually select .98 MP.
* Sam 3.1 latent mask for guided character replacement.
* I also added group bypass controls so no more errors.

fast forward to 2:42 to see sam 3.1 functionality

original post of workflow:  
[https://www.reddit.com/r/StableDiffusion/s/TaY5KEnFk7](https://www.reddit.com/r/StableDiffusion/s/TaY5KEnFk7)

edit: Posted v5.01

Fixed an issue where upscaler and interpolation weren't working with sam.

edit: posted v5.02

fixed an issue where bypassing audio 1 and video 1 caused errors. Fixed the thing i said i actually fixed in 5.01, for real this time.

edit: v5.03  
fixed the thing i said i fixed in 5.02. for real this time.

document fetched 2026-09-22 00:31 · canonical https://reddit.com/r/StableDiffusion/comments/1wm62zl/v5_update_newbie_minimax_h3_sam_31_latent_mask