Testing Minimax H3 locally in ComfyUI: multi-prop reference images and a perfect 5s seamless anime loop
Time-Ad-7720 · reddit · 2026-09-11
A hands-on test of the Minimax H3 reference-to-video model run locally in ComfyUI, targeting two challenges:
- Multi-prop single reference image: combining headphones, a Rubik's Cube, yo-yo, cassette and keyboard into one reference image instead of separate references, to test prop matching.
- Perfect seamless loop: a 5-second, 15fps scene in hand-drawn 1990s anime style with extremely subtle motion—gently swaying leaves, a few moving hair strands, slowly shifting sunlight—while the character and all objects stay perfectly still, and the final frame matches the first for an invisible loop.
Local generation took about 2 minutes. The post includes the full prompt (character/prop reference assignment, composition, lighting and animation constraints), a downloadable ComfyUI workflow file, and a matching Suno song link, making it fully reproducible.
More from Multimodal
- I tested 11 video upscaling methods for AI video: local, Topaz and cloud compared — lopas99 · 2026-10-28
- AI Horde adds Moss-Image for free, no-account image generation on volunteer GPUs — Wanabeelee2 · 2026-10-28
- AI band Techhalla reimagines Slipknot's 'Wait and Bleed' as a 1971 soul performance — Mother-Pear-2392 · 2026-10-28
- Inside the Hailuo H3 Max retro-MV workflow: 15-second chunks, one reference image, multi-shot prompts — techhalla · 2026-10-28
- Open-source pipeline turns Seedance AI video into real PBR textures on your 3D mesh — OdinLovis · 2026-10-28
- AI-generated cyberpunk short 'VHS Sirens' shows off consistent characters and neon city detail — brythehuman · 2026-10-28