MiniMax H3 Ref-to-Video: 15s Shot from 3 Images + Vocals, 1088p Latent Refinement
michael_goldenhour · reddit · 2026-10-18
A detailed workflow for MiniMax H3 reference-to-video with audio: 3 reference images, a 14.63s vocal track, and a 3,795-char structured prompt produce a 361-frame (15.04s) 1920x1088 shot with generated audio. Draft path: 12 steps at 960x544; quality path: 30 steps then 2x latent upscale plus a 10-step denoise pass at 1088p. The author notes refinement happens in latent space (not a resize) and that more denoising can degrade already-correct details; character identity and lip sync hold across the sequence.
Related event: MiniMax H3 Tested: 3 Reference Images Yield 15-Second MV with Vocals(2 posts)→
More from Multimodal
- 'Tesla in Pakistan' AI video made with Seedance 2.5 goes viral — SimplyAnnisa · 2026-10-18
- Borges's nine stories turned into walkable 3D worlds with AI model Fable 5.1 — emollick · 2026-10-18
- Midjourney cozy couple photo: full prompt shared for killing the AI plastic look — tisch_eins · 2026-10-18
- Higgsfield playbook: scene templates, paid creator network, $20-to-enterprise funnel — sujingshen · 2026-10-18
- Why did Anima flop? Illustrious still dominates the anime model scene — Front-Side-6346 · 2026-10-18
- AI-generated stage projections keep repeating "For a Brighter Tomorrow" — why? — Awkward-Concept5736 · 2026-10-18