AI NewsOpen sourceAnnouncement
Sam Yost released 3dicon, a Claude Code skill that generates a looping animated 3D icon from one prompt, with real transparency
Sam Yost released 3dicon on 23 September, an MIT-licensed Claude Code skill that takes one prompt and returns a looping animated 3D icon as an animated WebP with real transparency. The repository passed 300 stars in one day.
Image: GitHub
Why it mattersA designer or product team using Claude Code can now turn a short description into a fresh, on-brand looping icon file that drops straight into a UI, with a working alpha channel and a two-step approval before any paid video render.
Designers making a UI have picked either a real 3D pipeline or a static PNG for animated icons. Sam Yost released a Claude Code skill on 23 September that returns a looping animated 3D icon from a single prompt, delivered as an animated WebP with a working transparent alpha channel. The repository was published a day ago and passed 300 stars on 24 September under the MIT licence.
The skill installs through Claude's plugin marketplace, and a plain prompt like "make an animated 3d fire icon using /3dicon" starts it. It generates a still image first and pauses for the user's approval before it spends anything on the video step. It then generates a short video and pauses again. The two-step approval matters because the video model is the expensive part of the pipeline, and a still that the user does not like should never trigger a video render.
How the loop closes cleanly
Yost's own explanation of the trick is in the README diagram. The single generated still is sent to the video model as both the first frame and the last frame. The model returns to its own starting frame, and the loop closes with no visible seam. That constraint means the finished clip cycles cleanly instead of jumping between the end and the beginning, which is the common failure of an AI video used as a looping icon.
The transparency is treated as a separate problem. Instead of removing the background against an unknown colour, the pipeline runs the video against a specified background colour chosen by the tool. Because that colour is known, the alpha channel can be solved for exactly rather than guessed. That is what keeps soft edges soft in the finished WebP instead of leaving a coloured halo around the shape.
What powers it
The repository routes both the image model and the video model through OpenRouter, so a single API key covers the whole run. Yost lists GPT Image for the still and ByteDance's Seedance for the motion. FFmpeg has to be on PATH, and the first run downloads a matting model of about 180 megabytes.
The licence is MIT, and Yost says the icons a user generates carry no restrictions. That matters for a designer who wants to ship the output in a client's app. For anyone whose UI has been using a static PNG or a stock animation library, the practical result is that a fire icon, a spark, a globe or a padlock can be a fresh, on-brand looping asset from a description alone.
Source
- Repository: samyost1/3dicon
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.