FastH3 Ref2V: open-source controller for continuous character-driven AI video streams in ComfyUI
EarthDefenceForces · reddit · 2026-09-04
Building on the approach of jacokon/fasth3-live, the author has developed a browser-based Reference-to-Video continuous video stream controller for the MiniMax H3 model in ComfyUI, now open-sourced under GPL-3.0.
Features include: automatic injection of character reference images from a folder for continuous character-driven storytelling; a manual prompt queue with looping scenes; last-frame continuity between clips, enabling interactive storytelling during generation; runtime editing of LoRA, duration, prompts, and playback speed; adaptive quality based on video buffering; a custom music folder with sequential/shuffle playback; and an English/German bilingual UI.
More from Multimodal
- HKUST-GZ and Tencent open-source VibeWorlding, an agent that builds interactive 3D worlds from natural language — jiqizhixin · 2026-09-04
- Single-pass 18-second AI video with Wan2.2 int8 workflow, full params shared — r0ni · 2026-09-04
- Midjourney comic trick: give every color one job — coral for pressure, turquoise for the eye — tisch_eins · 2026-09-04
- Seedance 2.5 prompt recreates authentic early-2000s DV vlog footage — 'captured, not generated' — eyishazyer · 2026-09-04
- AI video of a city moving to shade one orchid — full prompt shared — umesh_ai · 2026-09-04
- Fair-fight video test: HappyHorse 1.1 beats Kling 3.0 on character consistency under identical prompts — kimmonismus · 2026-09-04