Developer Ryan has open-sourced Anatomy, an AI drawing skill that currently supports Claude Code, OpenCode, and Codex. Users can describe a technical concept in a single sentence, and the system will design a machine on its own, using moving parts to show how the concept works. The generated pieces are also interactive and can be controlled with a mouse or keyboard.
How Anatomy builds the visuals
According to the description, Anatomy first studies the relevant principles and parameters, then designs the parts, draws the graphics, and adds animation. The motion on screen and the numerical readouts are driven by the same computational model, so both update together during user interaction. It also includes automated checking tools to catch issues such as parts intersecting each other or visual occlusion errors.
33 featured works are already on display
One example is an explanation of an LRU cache, where the least recently used data is cleared first. For that case, the AI designs a bookshelf that moves books around. To explain how a mechanical watch works, it can generate a mechanical device that can be wound, speed-adjusted, and taken apart gear by gear. More complex examples include the Raptor rocket engine used on SpaceX Starship. That model can be broken into 11 groups of components and can also simulate ignition. The official site currently shows 33 main works.
Editable web output and reusable components
The final output is editable web code that can be opened as a standalone page or embedded into a website. Ryan has also packaged existing works into a component library so developers can use them directly. The project has been released under the MIT license.
Detailed builds still take significant time. Ryan said the mechanical watch took nearly two hours to make, while the Raptor rocket engine took more than 10 hours. The 3D feature is still in testing.


