Minimax H3 character swap tested: Turbo+VEDA hits 353s for 4-step generation
Mysterious_Pride_858 · reddit · 2026-10-14
A Reddit user benchmarked Minimax H3 character swap across acceleration schemes on an RTX 5090 + 96GB RAM, using the first 15 seconds of an 18s reference video at 1MP resolution.
- 4-step variants: Turbo LoRA 632s; Turbo+Sol 718s; Turbo+Spark-H3 682s; Turbo+VEDA fastest at 353s (322s sampling).
- 16-step single schemes: Sol 2310s; Spark-H3 2441s; VEDA 993s.
- Tests were driven via Codex Sol 6.1, so results may not be rigorous.
More from Multimodal
- Trying Out Fish Audio: "Kinda Cool, If You Ask Me" — sharpeye_wnl · 2026-10-14
- MiniMax H3 ecosystem roundup: ComfyUI 0.40, audio-first workflows, and style-consistency tips — optimisticalish · 2026-10-14
- Lightricks' APEX pairs hierarchical latents with async denoising for extreme 64-128x diffusion compression — Lightricks · 2026-10-14
- 6 open-source AI tools turn one sentence into a full short drama video — sven_ai · 2026-10-14
- China's AI short-drama flood backfired: viewer fatigue killed revenue and dragged down live-action too — TuhinChakr · 2026-10-14
- fofr shares his 3-step Argon workflow: Nano Banana refs, Lyria 3.5 music, TTS voiceover — fofrAI · 2026-10-14