huashu-motion turns a coding agent into a product launch film maker
huashu-motion is a new MIT-licensed Agent Skill that produces a 30-second product launch film from a brief, using GSAP motion graphics and composed music rather than video generation, with a point-and-click editor for frame-level fixes.
Image: GitHub
Why it mattersA founder or a small marketing team can now draft a product launch film by telling a coding agent what the product is, then edit the result by pointing at a frame rather than opening a video program.
huashu-motion is a new MIT-licensed Agent Skill that produces a 30-second product launch film from a one-line brief. Instead of calling a video generation model, it writes animated motion graphics with GSAP, Canvas, Three.js and SVG, composes original music from real instrument samples, and lays voice-over on top with Kokoro text-to-speech. The project is from a developer who goes by 花叔 and the repository has drawn 891 stars in its first day.
How a film gets made
The flow starts with one sentence. A user tells their coding agent, Claude Code or Codex, something like "make a 30-second launch film for my product". The agent calls the huashu-motion skill, which drafts three creative directions and asks the user to pick one. The chosen direction opens in a browser-based editor on the local machine. From there the user holds a key, clicks any scene, and describes what to change. The agent reads the point-and-click note, revises just that section, and shows the new output.
The pipeline uses local Chrome automation and FFmpeg to render the final MP4. The audio is CC0 or CC-BY instrument samples arranged by the skill, so a published film has no licensing problem. The output is a 30-second video with scored music, synchronised sound effects and optional narration. Everything runs on the user's machine.
Why motion graphics rather than video generation
The project description is careful about this. Video generation models are improving, and some teams use them to produce short reels, but they fail in the predictable places: a logo gets redrawn, a product shot drifts between frames, text comes out with letters that are not in the brand. A motion-graphics pipeline makes a different choice. The scenes are composed as code with real assets, so a logo stays identical from frame one to frame 900, text is text, and a product image is the product image the brand supplied. The visual range is narrower, and the output looks like motion design rather than cinema, but it is deterministic.
For a marketing team that was going to ask a designer to animate a static deck anyway, this is the shape of a usable tool. For a team that wanted a photorealistic film, this is not it.
Founders and small marketing teams
The audience is founders and small marketing teams who need launch material and do not have a motion designer on staff. The brief says the project has been tested with Claude Code (Opus), Claude Codex and other agents, including DeepSeek V4.1 Flash. The developer, 花叔, describes the work as inspired by Anthropic's Claude Motion feature, which the README references by name.
Three facts to weigh before adopting. The repository is one day old and has 891 stars, which is strong early signal, but no users have published a case study yet. The project runs on a user's machine and calls whatever coding agent the user is already paying for, so the real cost is a model session for each film, which has not been measured in the public release. And the output style is motion-graphics by design: a brand that needs a cinematic film is better served elsewhere.
For a solo founder who can describe a product in a sentence, the output is a working first draft of a launch film in minutes.
Source
The code is at alchaincyf/huashu-motion on GitHub.
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.