You may be familiar with Yann LeCunn's idea of JEPA and how it may be the real future of the artificial intelligence. I was reading the articles and watching his work on the topic and I was like it is one thing I could always use in my project of "SVG generation". Well, before that I used a model like FLUX or SD (finetuned on vector styles) and then used vtracer. Which is not really bad. But when I saw I-JEPA and how it behaves with images, I decided to give it a shot. So I made ...