# Minimax H3 Image's PE is too slow, are there alternatives or to use OAI-compatible API? — retained record

> Retained reddit record from reddit:StableDiffusion.

Canonical: https://highsignal.app/data/records/a7f87b8d660d77da

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:StableDiffusion)  published 2026-11-11 00:54 ingested 2026-11-12 02:40

[original source ↗](https://www.reddit.com/r/StableDiffusion/comments/1xt42b3/minimax_h3_images_pe_is_too_slow_are_there/)

## retained text



```
Title.

Currently, the image edit itself on 16MP took me 1min per image(4 step), which is fast.

But, with only 1 reference image, the PE takes 25 mins to generate 8k tokens, which is.. very slow.

Are there a better model to use(currently using the default template qwen3.5 9b w4a8 + abliterate lora) so its faster , or a way to integrate a OpenAI-compatible API here and uncensor the model? (The API have various models like gemini, glm, gpt, nemotron, qwen, etc.)
```



document fetched 2026-11-12 00:32 · canonical https://reddit.com/r/StableDiffusion/comments/1xt42b3/minimax_h3_images_pe_is_too_slow_are_there
