I trained a diffusion model (Flux Kontext LoRA) on caricature images, 3D-printed a Sharpie mount for my Ender 3, and hacked together a pipeline that goes photo → caricature → G-code. After a lot of trial and error it’s working pretty well, and I put up a little site where you can try it out.

Happy to answer questions or hear any feedback. Thanks!