MiniMax H3 Showdown: Which Looks Best? (132 Styles Tested)

I ran three MiniMax H3 video workflows on one RTX 4090: 132 animation-style prompts, 396 clips, and the same prompt and seed for each three-way comparison. From anime and claymation to paper, felt and experimental looks: which one is fastest, and which would you actually use? THE SPEED RESULTS Fast H3 Omni: 60 s average per 5 s clip; fastest on all 132 styles. H3 Turbo: 65 s average per clip. MiniMax H3 r2v: 190 s average per clip. Fast H3 Omni's mean render time was about 3.2x quicker than the tested H3 r2v workflow. These are local workflow timings, not a universal model ranking. Visual quality is yours to judge. CHAPTERS 00:00 The test: three H3 models, one RTX 4090 00:16 Broad animation styles 00:46 Clay, paper and graphic styles 01:56 Anime through the decades 03:16 Retro cartoon adventures 04:01 Folk tales and storybook worlds 05:01 Action and adult cartoons 06:21 Hybrid animation and dark fantasy 07:11 Illustration, crayon and felt 07:46 Technical and experimental looks 08:46 Retro digital graphics 09:06 Folk painting traditions 09:46 Abstract art and printmaking 10:36 Paper and textile crafts 11:16 Final scoreboard HOW I RAN IT One RTX 4090, local ComfyUI. 132 animation-style prompts x 3 workflows = 396 five-second clips, shown side by side. Each style has its own character and setting. The prompt and seed are identical across the three workflows for that style; the seed changes between styles. All three workflows used text-only prompts with no reference images: MiniMax H3 r2v: 8 steps with the VDN turbo adapter. Fast H3 Omni: 8-step sparse-attention distilled checkpoint. H3 Turbo T2V: 4-step turbo LoRA. Clips: 736 x 1280, 24 fps. Each prompt requests a close-up followed by a cut into motion. Workflow settings differ as listed above. Each model ran its whole batch serially. Times are server job start-to-finish measurements; first clips include model loading. Measured H3 r2v slowdowns are included in the averages. Total job time: 11 h 33 min; wall clock: 12 h 01 min. All 396 completed clips are included, with no retakes. This is a visual comparison with a measured speed scoreboard, not a scored visual-quality benchmark. The 132 styles were my selections from the earlier 196-candidate Flux 3 Klein test; some looks overlap. Presenter intro: Kai avatar animated with Seedance 2.5. The montage uses the project's original lo-fi instrumental music, not the models' generated audio. Which model would you use? Leave the style name, timestamp and your preferred panel. #MiniMaxH3 #AIVideo #ComfyUI