# Melting faces at 768x Minimax-H3 with audio sync for singing. Is there a solution? — retained record

> Retained reddit record from reddit:StableDiffusion.

Canonical: https://highsignal.app/data/records/e08057e9d9b20bb8

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:StableDiffusion)  published 2026-09-02 13:22 ingested 2026-09-03 02:41

[original source ↗](https://www.reddit.com/r/StableDiffusion/comments/1w5a8l5/melting_faces_at_768x_minimaxh3_with_audio_sync/)

## retained text



```
Greetings!

I find that whenever I do ref2va with audio sync (for singing) at 768x about half the time I get melting faces. (example posted) so I'm looking to fix but I don't have a decent pc so it's cloud or bust. I know it's something to do with low res pixels being stretched - don't really understant the tech but anyway

\-is that cause by my ref not being detailed enough?

\-assuming it's not really that - would the latest 2k 3H flagship model fix the low res issue (chatgpt reckons not)?

\-has anyone tried ComfyUI-H3-FaceRefine and how was it?

Thanks!
```



document fetched 2026-09-03 00:33 · canonical https://reddit.com/r/StableDiffusion/comments/1w5a8l5/melting_faces_at_768x_minimaxh3_with_audio_sync
