Testing MiniMax H3 music + lip sync locally: 25 steps minimum, 12 min per 15s clip on a 5090
dassiyu · reddit · 2026-08-21
A Reddit user shared hands-on experience running a MiniMax H3 music generation + lip sync workflow locally:
- A decent 15s clip needs at least 25 steps; using LoRA is not recommended;
- On an RTX 5090 in ComfyUI Kitchen, 15s at 0.9 takes about 12 minutes;
- H3's music generation is solid — noticeably better than the author's results with LTX;
- The main pain point is seamlessly stitching 15s segments; some claim 20s works, but the author's GPU nearly dies at that length.
More from Multimodal
- fal launches Worlds for real-time creation and interaction with AI-generated worlds — Scobleizer · 2026-10-16
- Gummy worms as a physics test: Gemini 4 Argon vs. Claude Opus 5.5 elasticity sims — DynamicWebPaige · 2026-10-16
- 260M-parameter looped diffusion transformer beats 6x larger baselines with 5x less compute — burkov · 2026-10-16
- Seedance 2.5 demo: stretching one second of an explosion into a 30-second shot — LudovicCreator · 2026-10-16
- Octopus Short Film Made with Grok Imagine, Upscaled via Topaz — creatoroff · 2026-10-16
- Open-sourced 61-second computing history film generated entirely by Muse Spark — sophiamyang · 2026-10-16