---
format: "aidr-story-markdown/v1"
id: "f89c73d39fcd277769c72d34104d1f5b1d9955b91e2e9cfa13d4f879b57ec94e"
canonical_url: "https://aidr.today/f89c73d3?lang=en"
title: "FastH3 v1 Generates 15 Second 768p Video in 13 Seconds in Open Weight Speed Leap"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-08-30T18:37:32.000Z"
category: "Models"
topics: ["nvidia","inference","video","chips"]
source_urls: ["https://huggingnews.com/ai/update-fasth3-v1-generates-15-second-768p-video-in-13-seconds-in-open-we-e31a7f06","https://x.com/NerdyRodent/status/2094032057936617727","https://x.com/shipfrontierai/status/2093796629522379166","https://x.com/FReza1984/status/2093813937771102599","https://x.com/AIfutureBenji/status/2094115268494819716"]
summary: "Hao AI Lab, NuvaLab, and Nvidia FastGen launched a high-speed video generation model based on MiniMax H3 that allows users to create content almost as fast as it plays. FastH3 v1 can produce 15 seconds of 768p video in 13 seconds using data-free DMD2 distillation. The model achieves a 14x speedup on Nvidia Blackwell B200 GPUs after more than 1,000 training hours. vLLM is working to integrate the system into vLLM-Omni, though independent testing is still needed to evaluate quality trade-offs."
---

# FastH3 v1 Generates 15 Second 768p Video in 13 Seconds in Open Weight Speed Leap

> [Open the canonical story](<https://aidr.today/f89c73d3?lang=en>)

**Published:** 2026-08-30T18:37:32.000Z
**Category:** Models
**Topics:** nvidia, inference, video, chips

## Summary

Hao AI Lab, NuvaLab, and Nvidia FastGen launched a high\-speed video generation model based on MiniMax H3 that allows users to create content almost as fast as it plays\. FastH3 v1 can produce 15 seconds of 768p video in 13 seconds using data\-free DMD2 distillation\. The model achieves a 14x speedup on Nvidia Blackwell B200 GPUs after more than 1,000 training hours\. vLLM is working to integrate the system into vLLM\-Omni, though independent testing is still needed to evaluate quality trade\-offs\.

## Sources

- [Story source](<https://huggingnews.com/ai/update-fasth3-v1-generates-15-second-768p-video-in-13-seconds-in-open-we-e31a7f06>)
- [Story source](<https://x.com/NerdyRodent/status/2094032057936617727>)
- [Supporting source](<https://x.com/shipfrontierai/status/2093796629522379166>)
- [Supporting source](<https://x.com/FReza1984/status/2093813937771102599>)
- [Supporting source](<https://x.com/AIfutureBenji/status/2094115268494819716>)