It's just an idea, but you should train the AI to "see" and analyze the uploaded image lines while zoomed out and zoomed in and follow completed lines while zoomed out, meaning that in the vector design it wont miss to close lines for a vector, because now sometimes it missess lines and where a line from a shape needs to be closed, its not completing it, creating a fault result and incompleted curves, I hope you get what I mean. Basically I want the vectorizer to complete the curves in the finished vector image. I'm not sure how to explain this... Sometimes for example hair - it needs to draw all of the hair, but its not completing some lines of that hair. So this is the zoom out/zoom in techniquie comes in handy for the AI to make the end result better.