Capy Video Gen Skill: Multi-shot AI video generation with face identity consistency. 300 experiments, 70% improvement. Script-to-video and idea-to-video pipelines for HappyCapy AI Gateway.
-
Updated
Mar 20, 2026 - Python
Capy Video Gen Skill: Multi-shot AI video generation with face identity consistency. 300 experiments, 70% improvement. Script-to-video and idea-to-video pipelines for HappyCapy AI Gateway.
The ComfyUI workflows, prompts, scripts and per-clip results behind a timing and face check of four ways to speed up MiniMax H3 on one RTX 4090, with the rule written before the numbers. Our own files only: no model weights, no LoRA files.
The ComfyUI graphs, settings, scores and per-seed tables behind a test of NVIDIA's Open H3 Regenerate 2K refiner on MiniMax H3 video of a synthetic test character: three refiner methods against upscalers, a rule written before the test, and a blind AI-tell read. Graphs, text and tables only: no model weights, no LoRA files.
Keeps your face and body yours in AI images. A Claude skill: one photo, a locked prompt, a canvas method for turned poses, and measured checks. Nano Banana, Gemini, Higgsfield. Free, MIT.
四张候选脸挑一张,出一张 16:9 资产表,角色可复用不撞脸 | Pick one of four candidate faces, ship one reusable 16:9 character sheet · 📮 [email protected] · 🎬 https://v.douyin.com/07EjIDGVErY/
To associate your repository with the face-consistency topic, visit your repo's landing page and select "manage topics."