---
format: "aidr-story-markdown/v1"
id: "b228101c3e9d88f964f3eda06ad7024b5a192a11a1df8b06c92b675ac655f7a3"
canonical_url: "https://aidr.today/b228101c?lang=en"
title: "Show HN: We compressed a 31B video+audio diffusion model to 14B, 6 steps, INT8"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-11-07T15:16:01.000Z"
category: "Open Source"
topics: ["inference","open-source"]
source_urls: ["https://github.com/Video-Rebirth/hyperflow-h3-flash","https://news.ycombinator.com/item?id=50405263"]
summary: "HyperFlow H3 Flash is a 13.7B compressed MiniMax-H3 model for joint video and audio generation using 4- to 8-steps in INT8 and GGUF formats."
---

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

> [Open the canonical story](<https://aidr.today/b228101c?lang=en>)

**Published:** 2026-11-07T15:16:01.000Z
**Category:** Open Source
**Topics:** inference, open\-source

## Summary

HyperFlow H3 Flash is a 13\.7B compressed MiniMax\-H3 model for joint video and audio generation using 4\- to 8\-steps in INT8 and GGUF formats\.

## Sources

- [Story source](<https://github.com/Video-Rebirth/hyperflow-h3-flash>)
- [Discussion](<https://news.ycombinator.com/item?id=50405263>)