# HN: Show HN: We compressed a 31B video+audio diffusion model to 14B, 6 steps, INT8 — retained record

> Retained hackernews record from hackernews.

Canonical: https://highsignal.app/data/records/4f7806bbf22fff2b

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


source: [hackernews](https://highsignal.app/data/hackernews) published 2026-11-07 15:16 ingested 2026-11-08 02:43

[original source ↗](https://news.ycombinator.com/item?id=50405263)

## extract

Points: 58 | Comments: 12
Author: jiant
Link: https://github.com/Video-Rebirth/hyperflow-h3-flash
Show HN: We compressed a 31B video+audio diffusion model to 14B, 6 steps, INT8

## retained text



```
Points: 84 | Comments: 14
Author: jiant
Link: https://github.com/Video-Rebirth/hyperflow-h3-flash
Show HN: We compressed a 31B video+audio diffusion model to 14B, 6 steps, INT8
```



document fetched 2026-11-14 02:59
