Open-Source ComfyUI Studio for Long-Form AI Video with MiniMax H3 Character Consistency
mnm9678 · reddit · 2026-09-09
A developer released qamba-studio v0.2.3 (open-source macOS/Windows builds), a local-first desktop app for long-form AI video built on ComfyUI and MiniMax H3.
Key design points:
- Storyboard & bible system: characters, locations, and props get reference sheets (face, full body, turnaround rendered as one continuous H3 pass), and every shot stages from these sheets so faces and places stay consistent across an episode
- One-shot wizard: give it a brief and it plans story, scenes, shots, cast, and references, then queues renders with no manual setup
- Director chat: an in-app agent can rewrite scenes, re-render blocks, and edit the project mid-flight; uncensored option available
- Motion continuation: chained shots pin the previous block's tail frames and audio into the next render so movement doesn't reset at cuts
- Post-processing chain: SeedVR2 restore/upscale, LTX 2.5 detail pass, FILM/RIFE 4.26 interpolation, H3 FaceRefine, and color grading—chainable in any order
- Auto-installs ComfyUI or reuses an existing one; model catalog checks file size and VRAM before download. Supports i2v/t2v/flf/r2v int8-quantized H3, with LTX 2.5 and Wan 2.2 for reference/still work
Feedback welcome; demo video attached.
More from Multimodal
- Creator Combines Grok Imagine Images with MiniMax H3 for Portal Video Demo — creatoroff · 2026-09-09
- Midjourney blends Arabesque and Origami styles into striking Islamic paper-art patterns — aziz4ai · 2026-09-09
- Creator burns all Magnific CPP credits in 3 days making AI series trailer 'DAKOTA' — gen_ericai · 2026-09-09
- Palmier pits its AI editor against DaVinci's new MCP: 3 min vs 10 min on the same edit — _AustinCalvert_ · 2026-09-09
- AI Nature Documentary Narrates Open-Plan Office Humans Like Wildlife — Pale_Coyote7451 · 2026-09-09
- Cosmos3 (64B) INT4 Quants Bring Local Image and Video Gen to Mac and CUDA — Formal-Swordfish-228 · 2026-09-09