Open-Source Ref2V Workflow Auto-Transcribes Media and Formats Prompts for H3 Video Generation
bstr3k · reddit · 2026-09-10
A Redditor spent 2.5 weeks building a beginner-friendly H3 Ref2V workflow (on Hugging Face) with two custom nodes: one transcribes/captions your reference video and images, another loads a small LLM of your choice to write properly formatted Ref2V prompts from your intent. The full workflow generates the video too.
Hard-won tips included:
- On low VRAM, run the prompt enhancer separately and review output before generating
- H3's context window appears tied to system resources; exceeding it garbles audio/motion—lower output length and resolution
- Specify emotions explicitly or you get expressionless characters: "slop prompt in, slop video out"
- When the replacement person looks too similar to the original, the model converges back—swap via a faceless green intermediate person or use Sam3 character replacement
- Describe multi-person scenes specifically; "replace the man wearing white shorts with the man in <picture 1>" beats vague prompts
- Adding one-line scene context for the LLM (e.g. "<video 1> is a video of two girls eating chocolate ice cream") helps a lot
Related event: Open-source Minimax H3 video workflows gain traction(2 posts)→
More from coding & agent
- LangChain ships Managed Deep Agents 0.9, walkthrough from Interrupt London keynote — LangChain · 2026-10-26
- Knowledge Worker Kit adds a full section on organizing projects — dSebastien · 2026-10-26
- Spring AI 2.1.0-RC1 adds tool-call confirmation, pausing agent loops before real-data access — therealdanvega · 2026-10-26
- METR red-teams OpenAI's agent monitoring, finds bypasses and gaps — thione · 2026-10-26
- OpenAI: coding agents weakened safety monitoring, one lied about it — thione · 2026-10-26
- ConvFill: A 0.6B Talker Buys Time While a Large Reasoner Whispers the Real Answer — andimarafioti · 2026-10-26