20 drafts later: a full local MiniMax H3 pipeline for an EVE Online AMV, notes and prompts included
DienerTech · reddit · 2026-10-04
Over a month, 20 complete drafts, and hundreds of generations, the author produced an EVE Online PvP-style AMV, 'Can You See Me Now,' running MiniMax H3 locally — and open-sourced notes, prompts (failures included), and workflow guides at github.com/dienertech/lab-notes.
Pipeline
- MiniMaxH3ReferenceToVideo + MiniMaxH3AddGuide in ComfyUI, 960×544, 24fps, 20 steps, resmultistep/beta, upscaled to 1440p with plain Lanczos (no AI upscaler)
- Reference and storyboard images via GPT Image; no difference found between Codex's GPT Images tool and direct API generation
- Claude Code and Codex served as production assistants throughout
Key lessons
- The reference image is the script: H3 faithfully animates mistakes in the still (bent hull, extra gun pod); no prompt can salvage it
- Guide frames matter: first-frame-only drifts on long takes; first + last (last edited from first) bounds actions; adding an interior guide on 158-frame takes held 4/4 clean vs 2/7 without
- Negative prompts don't stop background floods (invented planets, weird glows); changing seed or reference does
Next up: similar videos for a D&D campaign, with humanoid animation as the new challenge. A solid distillation of community H3 research.
More from Multimodal
- invideo Shares Workflow for Keeping Film Locations Consistent Across Scenes — aziz4ai · 2026-10-04
- Sholto Douglas: We're witnessing a new AI artform and artist being created live — _sholtodouglas · 2026-10-04
- Keeping locations consistent in MiniMax H3: a 20-photo grid reference trick — PATATAJEC · 2026-10-04
- AI-Generated Fantasy Short 'The Sundered Light': Three Fairy Sisters, One Broken Seal — Zigg-Zagg-PiX · 2026-10-04
- FaceFusion v4 Beta ships as its biggest release yet, with $1000 challenge — henryruhs · 2026-10-04
- Rebuilding an After Effects workflow in Python with a Claude×GPT relay, free MIT rig released — AnonymousClaudeuser · 2026-10-04