# looking for an LLM for tasks like h3 reference mode prompting that runs on my 8gb 3060ti and 32gb ram — retained record

> Retained reddit record from reddit:StableDiffusion.

Canonical: https://highsignal.app/data/records/37a4d6c6a86c18f9

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:StableDiffusion)  published 2026-09-21 16:40 ingested 2026-09-22 02:38

[original source ↗](https://www.reddit.com/r/StableDiffusion/comments/1wmi0zr/looking_for_an_llm_for_tasks_like_h3_reference/)

## retained text



```
What I do at the moment is ask Grok to rewrite a casual prompt to follow this formatting guide: [https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/docs/VIDEO\_PROMPT\_WRITING\_GUIDE\_ref\_en.md](https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/docs/VIDEO_PROMPT_WRITING_GUIDE_ref_en.md), and it’s working great.

I'm just wondering if such a task can be done locally. I'm not expecting it to run super fast or anything; just 5 to 10 minutes for a task is fine.
```



document fetched 2026-09-22 00:31 · canonical https://reddit.com/r/StableDiffusion/comments/1wmi0zr/looking_for_an_llm_for_tasks_like_h3_reference
