video-talkcraft

video-talkcraft is an open-source Agent Skill by Vincentwei1021, the voiceover-driven explainer part of the video-shotcraft series. Give it a narration script and a finished voiceover, and it aligns word-level timestamps on your machine, storyboards every beat into a SHOTBOOK, then renders a polished explainer with Remotion: kinetic type, evidence screenshots, camera moves, plain-cut subtitles and film-grade sound effects, all locked to the voice. What it does - Aligns the voiceover to the script at word level, locally, so captions and motion sit on the voice - Storyboards each semantic beat into a SHOTBOOK, drawing on 108 motion recipe cards - Follows a seven-layer, anti-slideshow shot system that keeps camera moves slow and motion restrained - Films web pages with a camera (scroll, tour, magnifier) instead of pasting static screenshots - Opens a workbench, a timeline post-production desk, after delivery, where the 108 cards are parameterized and clips export with an alpha channel for CapCut, Premiere or After Effects How it installs - It is an Agent Skill for Claude Code or Codex. Clone the repository and run its runtime setup (bash runtime/check-runtime.sh --upgrade), which installs the shared Remotion dependencies and headless browser. - Needs Node.js and Remotion. A single shared node_modules is about 760 MB. Good to know - Inputs are a narration script and a voiceover, plus any character footage. Digital-human generation is not part of the skill; character footage is an input. Fish Audio can optionally synthesize the voiceover and timestamps from the script. - The methodology docs and recipe cards are written in Chinese; the toolkit is Chinese-narration-first, and mixed Chinese and English narration is supported. - Licensed under PolyForm Noncommercial, for non-commercial use.