Tell the agent "Make a 90-second documentary about the Wright brothers," and it will research facts, verify historical accuracy, write a storyboard, first synchronize timing using system voice, then find public domain photos from Wikimedia Commons and record their licenses, use Python's Skia to arrange photos, clippings, documents, maps, and counters within the same designed scene, trigger stamps, highlighters, and number rolls when specific words are spoken in the narration, and finally synthesize sound effects, optional background music, and burned-in subtitles. There are four styles: evidence wall, product launch, blueprint, and night scene; the agent will pause for your confirmation at three points: the storyboard, visual style, and a 30-second draft. The README states that rendering a 92-second 1080p final video on a laptop with 16 GB of RAM takes approximately 6 minutes.
Why it matters The visuals are drawn frame-by-frame by code, not generated video, ensuring consistent results for every render; if you swap the voiceover, the entire film automatically re-synchronizes to the new audio.
Stars 882 (+34 / 24h)
Reasons
- 34 stars added in 24 hours, total stars 882 (indicating attention only); MIT license
- Images default to files with clear licenses from Wikimedia Commons, rejecting non-free licenses, and automatically generate credits.md; sound effects are synthesized by code, so no additional licensing is needed
Usage
- Run npx skills add Moh4696/cinematic-skill (requires Node 22.20 or higher), then pip install -r requirements.txt to install skia-python, numpy, and pillow, ensure ffmpeg is installed, start a new session, and tell the agent "Make a 90-second documentary about ××"; also available as plugins for Claude Code and Codex, and an extension for Gemini CLI
Risks
- The only example provided is the author's own short film about the Wright brothers; you are responsible for copyright when using your own photos or AI-generated stills
- You must provide the final voiceover via TTS or human recording yourself; system voice is used only for timing synchronization