File size: 1,053 Bytes
66d260c
 
 
 
7f08972
552b364
7f08972
66d260c
 
32aa931
69ea65f
 
9de2b7a
69ea65f
9de2b7a
69ea65f
9de2b7a
69ea65f
9de2b7a
69ea65f
66d260c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
This is just an example of using MiniMax H3 as an image editor. The actual workflow that I use requires several custom nodes, some of which are not published, so this one is simply a bare bones demonstration.

This uses the hybrid MiniMax H3 model from here: https://huggingface.co/smhfacct/Minimax-H3-fl2va-ref2va-hybrid-models/tree/main

It uses the custom VAE from here: https://huggingface.co/Mamad8/MiniMax-H3-Image-VAE/tree/main

It uses the LoRA from here: https://huggingface.co/Kijai/MiniMax-H3-experimental/tree/main/loras 

To run MinMax H3 for a single frame, you need to patch "ComfyUI\comfy_extras\nodes_minimax_h3.py". Instructions for this can be found here: https://huggingface.co/Mamad8/MiniMax-H3-Image-VAE/discussions/2#6a78478e7201a2a765aded0c
(You just delete the code in nodes_minimax_h3.py and replace it with the code in that comment. It changes nothing other than reducing the minimum frame count from 5 to 1)

It uses these custom nodes:

rgthree-comfy

ComfyUI-KJNodes

ComfyUI-VideoHelperSuite

ComfyUI-Spectrum-MiniMax-H3