# Teeth, Cheeks and Eyes - LTX-2.5 — retained record

> Retained reddit record from reddit:StableDiffusion.

Canonical: https://highsignal.app/data/records/0794128c9fa9060a

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:StableDiffusion)  published 2026-09-20 08:33 ingested 2026-09-21 02:37

[original source ↗](https://www.reddit.com/r/StableDiffusion/comments/1wlbjqq/teeth_cheeks_and_eyes_ltx25/)

## retained text



```
Hey everyone, 

I am having this issue... I am creating marketing videos for work using LTX-2,5 but I am struggling to have a good speaking avatar... I created a setup photo using Z-Image-Turbo, adjusted it via Flux Klein to add specific needed details to the photo and I am trying to let my model explain a specific topic.

the issue I am having, although I am generating 1080p videos, and maxmum 10 seconds per generation, I am having artefacts such as bad teeth (grainy), or bad cheek movements or blurry messed up eyes (attached picture).

I tried bumping up the quality above 1080p, I tried facial reconstruction, I tried upscaling... though I did not find yet the ultimate formula... has anyone collected experience around that field? does anyone have a working workflow for such videos?

I am open for your input and recommendations

thank you
```



document fetched 2026-09-21 00:30 · canonical https://reddit.com/r/StableDiffusion/comments/1wlbjqq/teeth_cheeks_and_eyes_ltx25
