3 months with MiniMax H3: 100 videos published, single generations hit 40 seconds
unluigui · reddit · 2026-11-15
A Redditor recounts three months of working with the MiniMax H3 model: starting in August running it locally in ComfyUI, they've now published 100+ singing videos on Instagram and passed the milestone days ago. Single-run generations now reach 40-42 seconds with generated audio. Work is shared on Civitai; longer singing videos drop on weekends only since they take long to prepare, with shorter clips on weekdays.
More from Multimodal
- Dreamy purple 'portal opens at night' AI image series catches attention — Kyrannio · 2026-11-15
- One Kyoto photo becomes a walkable 3D world via Atlas, Three.js and Claude — Scobleizer · 2026-11-15
- Custom Image Model Meets VR Gaussian Splatting in Eerie Demo — darlens13 · 2026-11-15
- Oil-painting-style AI-generated Sailor Moon art wows X with classical texture — Kyrannio · 2026-11-15
- Fable 5.2 renders every single flower in the scene as true 3D — Kyrannio · 2026-11-15
- Claude-built Blender spider film: 105M tokens, ~20 prompts, $2.21 GPU render — Gambo7592 · 2026-11-15