# Resistance Sci-Fi - MiniMax H3 — retained record

> Retained reddit record from reddit:StableDiffusion.

Canonical: https://highsignal.app/data/records/e7f22ebe95313fe6

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:StableDiffusion)  published 2026-10-02 15:08 ingested 2026-10-03 02:38

[original source ↗](https://www.reddit.com/r/StableDiffusion/comments/1wvw97s/resistance_scifi_minimax_h3/)

## retained text



```
This is a short film I made to push the limits of MiniMax H3 on ComfyUI, mainly using the Hybrid model.

It took me about three weeks to complete, running locally for most of the work, with some generations done on RunningHub using a modified workflow by Pixaroma.

I used ChatGPT mainly to help create prompts for the image generation in Nano Banana, but I wrote all the video prompts myself. There’s a reason I try not to rely on AI for video prompting: I want to understand how H3 actually works. 

One thing I’ve found is that H3 doesn’t necessarily need the official prompting structure. In some cases, that structure can actually make things more complicated than they need to be.

SeeDance 2 fast was used in the kiler drone attacking the men scene, since this is something beyond H3.

The music was not AI-generated. It’s from my own library collection.

Edited in DaVinci Resolve 21.

Hope you enjoy this rather long piece of work, and I’d really appreciate any feedback, especially from anyone experimenting with H3.
```



document fetched 2026-10-03 00:32 · canonical https://reddit.com/r/StableDiffusion/comments/1wvw97s/resistance_scifi_minimax_h3
