# Rapid-H3: Super fast H3 video + audio in about 2 minutes on one DGX Spark — retained record

> Retained reddit record from reddit:StableDiffusion.

Canonical: https://highsignal.app/data/records/858ab800ad954cad

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:StableDiffusion)  published 2026-10-30 06:36 ingested 2026-10-31 02:38

[original source ↗](https://www.reddit.com/r/StableDiffusion/comments/1xj824f/rapidh3_super_fast_h3_video_audio_in_about_2/)

## retained text



```
Rapid-H3 makes MiniMax-H3 (text / image → video with stereo audio) render **3–5× faster on a single NVIDIA DGX Spark (GB10)** than stock ComfyUI, with no human-noticeable loss in picture, motion, voice or lip sync. It is a ComfyUI node pack plus two ready-made render presets:

* `ts_h2` (default): **≈117 s** for a 768×1344, 141-frame (≈5.9 s) first-frame clip with audio.
* `ts_s6` (camera moves, close-up faces): **≈201 s**, at stock level on every criterion we score.
* Stock ComfyUI bf16 at the 10-step reference: **590–630 s**.
* The first render is slower on a fresh install, while the CUDA kernels are built (once) and the models load.
```



document fetched 2026-10-31 00:31 · canonical https://reddit.com/r/StableDiffusion/comments/1xj824f/rapidh3_super_fast_h3_video_audio_in_about_2
