~ / blog / series / MiniMax-H3 on RTX 5090
❯ ls ~/blog/series/minimax-h3-on-rtx-5090
15 posts
- partdatetitle
- 12026-08-04[Benchmark] Running MiniMax-H3 on one RTX 5090: four files, 31.7 GB, 175s per talking clip
Beginner walkthrough for MiniMax-H3, the 33B model that generates video and stereo audio in one pass. Full precision is 115 GB; quantized it fits one RTX 5090. What to download, where it goes, how to prompt it.
- 22026-08-06[Benchmark] Twice as Fast: MiniMax-H3 on an RTX 5090, 625s Down to 314s
Three stacked changes took a 15-second 1080p MiniMax-H3 render from 625s to 314s on one RTX 5090: 14 steps, SageAttention 2.2.0, and RTX VSR replacing Real-ESRGAN.
- 32026-08-23[Benchmark] MiniMax-H3 1080p Isn't Unsupported, It's Three and a Half Hours
Three mistakes shooting a wuxia scene in MiniMax-H3: a 502 that wasn't a rejection, blur upscaling can't fix, and a prompt that described a face instead of naming one.
- 42026-08-31[Benchmark] Ten effect embeddings for MiniMax-H3 — 10 MB that buys you bullet time, fire breath and a full year of seasons
Ten community effect embeddings for MiniMax-H3, tested at native 1344x768 on one RTX 5090. What each one actually does, the prompt shape that lets them work, and the placement rule that decides whether they fire at all.
- 52026-09-03[Benchmark] From OOM to 8.6 minutes: a MiniMax-H3 config stack for one RTX 5090
MiniMax-H3 at native 1344x768 on a single RTX 5090: 15 seconds of video in 518 s. What KJNodes chunking, torch cu130 and the NVFP4 kernels are each worth.
- 62026-09-06[Benchmark] Two characters in one shot: MiniMax-H3 Ref2VA takes multiple reference images
Ref2VA is the only MiniMax-H3 mode that takes several reference images. Two characters, 243 frames, 437 s on one RTX 5090, and CER 6.8% on the dialogue.
- 72026-09-11[Benchmark] LTX-2.5 vs MiniMax-H3 on one RTX 5090: 29s vs 81s for the same Chinese dialogue clip
LTX-2.5 and MiniMax-H3 on one RTX 5090, same Simplified Chinese line: 28.67s vs 81.22s end to end, both at CER 0%. Files, params, VRAM ceiling.
- 82026-09-13[Benchmark] Roll cheap, then enhance the keeper: NVIDIA's two-stage MiniMax-H3 + LTX-2.5 pipeline explained
NVIDIA's Sol-H3 two-stage recipe on an RTX 5090: MiniMax-H3 drafts at 672×384 in 25.5 s, LTX-2.5 refines only the take you keep. Downloadable ComfyUI workflow.
- 92026-10-12[AI Workflow] Block a scene in Blender, let MiniMax-H3 film it: local Ref2VA on an RTX 5090
Block out a shot with capsules and boxes in Blender, feed it to MiniMax-H3 Ref2VA with two reference photos, and get a live-action clip that keeps left/right order and occlusion. 89.9 s on one RTX 5090.
- 102026-10-16A 22-second Taiwan FPV flight without a drone: Blender blockout + MiniMax-H3 on one RTX 5090
A 22.7-second FPV flight that reads as one take, over Yushan, Taroko Gorge, Taipei 101 and Sun Moon Lake, built from a gray Blender blockout and MiniMax-H3 Ref2VA on one RTX 5090.
- 112026-10-18A 25-second no-fog FPV one-take with MiniMax-H3 and ComfyUI-H3-Continuum on one RTX 5090
A 25-second FPV one-take over Taiwan with no fog: MiniMax-H3 Ref2VA and ComfyUI-H3-Continuum on one RTX 5090, each splice hidden in a turn through the sky.
- 122026-10-19Turn a real dance into a depth video and have MiniMax-H3 restage it in a palace: native 768 on one RTX 5090
Real dance to grey depth video to MiniMax-H3 Ref2VA with a first-frame guide: a new character dances the routine in a palace, native 768x1376 on one RTX 5090.
- 132026-10-25[MiniMax-H3] A storyboard image isn't a prompt: getting local H3 to hit 13 of 16 shots
A 4x4 storyboard plus a character sheet into MiniMax-H3 Ref2VA on an RTX 5090: one sentence hit 3/16 panels, an official-format shot timeline hit 13/16.
- 142026-10-30[MiniMax-H3] Open-H3-Regenerate-2K on an RTX 5090: more detail than VSR at 133 s a clip
NVlabs' Open-H3-Regenerate-2K on one RTX 5090: 133 s per clip vs 19 s for RTX VSR, real eyelash detail, and the cost: faces drift and small sign text turns into fake characters.
- 152026-11-01Making a character do parkour with NVIDIA MotionBricks on one RTX 5090: 1.66 ms per generation
NVIDIA MotionBricks on an RTX 5090: 1.66 ms per generation, a 10-jump parkour run rendered in Blender, then used as the skeleton for a 20 s MiniMax-H3 video.