I considered running the vectorization before showing any of the elements to the user, so the user would be editing everything as a vector. I didn't go with this approach since the results of the vectorization weren't consistently good for all the images. But I'm aware that the vectorization step is the step that requires more work, the problem is that it already has received most of it.
The main options for the vecotrization are:
- Straighter lines / smoother curvy lines
- Less noise / more details [Advanced]
- Number of colours: simpler vector / more details [Basic]
I even looked for deep learning approaches to vectorization but they tool a really long time to run: https://arxiv.org/abs/2306.06441