Claude P5.js Drawings
![Claude P5.js Drawings](/content/images/size/w2000/2024/06/beans.png)
I'm checking in on LLMs' ability to draw things using P5.js. It seems like things have improved a lot since the last time I tried some of these same experiments in 2022 (post). I used claude.ai to generate p5.js code, then copied the result into the p5.js playground and grabbed screenshots. The model must have a deep understanding of the context for these scenes, and how to break them down into basic shapes and position things correctly relative to each other.
![](https://danielpquinn.com/content/images/2024/06/image-1.png)
![](https://danielpquinn.com/content/images/2024/06/image-2.png)
![](https://danielpquinn.com/content/images/2024/06/image-3.png)
![](https://danielpquinn.com/content/images/2024/06/image-4.png)
![](https://danielpquinn.com/content/images/2024/06/image-5.png)
I asked Claude to reproduce a low poly 3D model using p5.js, and I'm impressed by how close it was able to get. I really don't understand how this is possible 😥.
![](https://danielpquinn.com/content/images/2024/06/image-6.png)