# MiniMax H3 Image 20MP Workflow — retained record

> Retained reddit record from reddit:StableDiffusion.

Canonical: https://highsignal.app/data/records/180493f32d317cd5

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:StableDiffusion)  published 2026-11-11 05:36 ingested 2026-11-12 02:40

[original source ↗](https://www.reddit.com/r/StableDiffusion/comments/1xt9ocv/minimax_h3_image_20mp_workflow/)

## retained text



```
The official ComfyUI workflows for MiniMax Image are pretty bad at showcasing the model's capabilities. These sampler/scheduler settings are from Kijai's Minimax testing in discord.  
  
There is still some work needed to improve fine details (possibly fixable by loras?) but overall the outputs look pretty good!

The workflow has T2I and Edit modes, just enable the one needed via the bypass node or bypass the group manually.

The sampler/scheduler is the SamplerLCM node + Simple 6 steps with  
s\_noise: 1.10  
s\_noise\_end: 0.90  


[Link to the workflow](https://huggingface.co/buckets/n8te0/diffusion/tree/Minimax-H3-Image-20MP/Minimax-H3-Image-20MP.json)  
  
[Original Output Images](https://huggingface.co/buckets/n8te0/diffusion/tree/Minimax-H3-Image-20MP)
```



document fetched 2026-11-12 00:32 · canonical https://reddit.com/r/StableDiffusion/comments/1xt9ocv/minimax_h3_image_20mp_workflow
