topic-to-video
Each film began as one request in a chat. topic-to-video researched the subject, wrote and voiced the script, designed every shot and rendered the video. No frame was animated by hand.
Each shot is a React component rendered frame by frame with Remotion, picked from a library of 225 shot designs or built for the film. Heavier 3D shots are rendered in Blender.
Tencent WorkBuddy’s desktop app rebuilt from its official screenshots, with six shots rendered in Blender: the desk companion, a pile of files, a MacBook Air, and the app itself on a MacBook screen. Unofficial, not affiliated with Tencent.
The skill’s own launch film. It plays real clips from other films on this page and shows the request each one started from.
A launch film for slide-maker, cut to its music. Every deck page on screen is a real page the tool made, not a mock-up.
What Grok Bot does and where it stops, in 36 beats under three minutes. An independent explainer, not affiliated with xAI.
Sound as squeezed and stretched air, the flipped wave that cancels it, the two microphones in an earbud and the tiny window of time they have to work in.
A promo for the job-hunt skill: finding roles, judging fit, tailoring a CV and rehearsing interviews. The CV and reports on screen are a fictional example.
Each beat is a web page built for the film, animated in headless Chrome and captured frame by frame. Chrome is all it needs.
Why tearing off one corner still scans and tearing the wrong spot does not. Every code on screen was generated for the film and checked with an independent decoder.
An MRI scanner never takes a photograph: precession, relaxation, gradients and k-space. The k-space on screen is a Fourier transform computed for the film.
Install the skill in Claude Code, then ask it for a video about your topic.
git clone --depth 1 https://github.com/addsumtech/topic-to-video.git ~/.claude/skills/topic-to-video