claude-motion lets coding agents produce launch videos with their own sound
An open-source toolkit called claude-motion teaches a coding agent motion design rules, a self-review loop over rendered frames and a Python-stdlib sound engine, so the agent can produce short videos without After Effects or stock audio.
Image: GitHub
Why it mattersA marketer or product team can ask a coding agent for a short launch video or explainer and get a numbered motion plan with its own soundtrack, instead of writing a brief for a motion designer and waiting.
A toolkit called claude-motion lets a coding agent produce short motion graphic videos end-to-end: taste-level rules with specific numbers, a review loop that checks its own rendered frames and a sound engine written in the Python standard library. The repository, at whaleyxbt/claude-motion under an MIT licence, reached 467 GitHub stars in the 19 days since it was published.
The maintainers position it as a motion design toolkit for coding agents. It ships Claude Code and Cursor skills in .claude/skills/, and works with Codex, OpenCode or any agent that reads AGENTS.md. The README notes that Remotion skills elsewhere teach the agent the API; this one teaches it taste, verification and sound.
The three pieces
The README describes motion-design rules written as numbers rather than adjectives. Entrance easing is written as bezier(0.16, 1, 0.3, 1). Exits are half the duration. The stagger between words in a word-by-word reveal is 0.09 seconds. Reading holds, type scale, one-accent colour, texture and motivated transitions are specified the same way, with numbers an agent can act on.
The self-review loop is the toolkit's answer to the fact that agents cannot watch video. The agent renders a draft, gets back a contact sheet with one timestamped frame per beat, checks the frames against a list that covers collisions, clipping, holds, safe area and phone readability, and fixes what it finds before declaring the piece done.
The sound engine is pure Python standard library, so there are no samples, no licence fees and no third-party dependencies. The README lists generators for typing, clicks, pops, whooshes, risers, chimes, impacts, textures and pads, a mixer that places sounds on the same timeline as the picture, and mastering that hits negative 14 LUFS by itself. The agent writes one cue sheet per video and adds generators when it needs a new sound.
A second pipeline for web simulations
The repository ships a second pipeline for pieces that are simulations rather than scripted graphics. A single HTML page carries the whole simulation, with three.js and fonts inlined so the file works offline. The simulation is a pure function of time, with physics and agent state computed once at load so any frame can be drawn on demand. Headless Chromium steps the page frame by frame into ffmpeg, which captures 60 frames per second at any size regardless of how slow a single frame is.
The maintainer notes a related tool, beats.py, that finds the tempo, bars and drop of a music track and hands them to the simulation so each cut happens on a beat. Three example simulations ship with the repository: a wall of 256 agents crawling websites, a username rendered into a dossier with red threads, and a camera flight through a running rocket engine with sound generated in Python.
The claim this toolkit makes is modest. It names the frame rate, the resolution and the format of the output, and leaves the question of whether the result reads as a professional motion graphic to the viewer. For a team briefing a motion designer this week, a one-prompt first draft is a cheaper way to find out whether the idea is worth paying for a human to finish.
Source
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.