112 MiniMax H3 Tests on an RTX 3090: Hunting the Golden Settings
badincite · reddit · 2026-09-07
The author ran a 112-case quality sweep of MiniMax H3 on an RTX 3090 (32GB RAM, CUDA 13.0, ComfyUI 0.34.4), holding prompt, seed, reference image, Sage setup, sampler and scheduler constant while varying model/LoRA combos (C1–C7, pruned/int8/turbo variants on Hugging Face), native resolution (0.4–0.98 MP), and steps (6–12).
Best quality per resolution: 0.4 MP → C4 at 12 steps (246s); 0.6 MP → C3 at 12 steps (271s); 0.8 MP → C3 at 10 steps (256s); 0.98 MP → C3 at 12 steps (337s). Best speed/quality balance: C3 at 0.8 MP / 6 steps (166s) but with poor audio; 8 steps (210s) turned out best overall. C3's pruned 8-step turbo variant emerges as the strongest general pick.
More from Multimodal
- LichtFeld Studio v0.5.4.3 ships community-built SteadySplats, POPSpa and Voxelization — janusch_patas · 2026-10-26
- One prompt to a 2-3 min narrated short: hands-on test of Antigravity video generation — AI_Andrew · 2026-10-26
- Creator makes 10,564-frame animated short entirely with Gemini toolchain — AI_Andrew · 2026-10-26
- FLUX.3 Klein step-distilled to 4-8 steps with CFG baked in — Fit-Structure-6489 · 2026-10-26
- Full animation short made with Google Gemini tools, 10,564 frames — AI_Andrew · 2026-10-26
- ostris pre-training "Primal VAE": 1-channel latents match modern 8x VAEs via patching — ostrisai · 2026-10-26