A comment about suggestion 1: The creators may have prior computer vision experience, but as someone who tried to learn this from scratch and perform handwritten shape detection I can say that it's a messy place to work. If it's not their core competency, it can get tricky with all the different shapes and problems that may arise with said shapes. Then again, I didn't have machine-learning experience which may be the way to go.

That would definitely be cool, though. You could basically edit what you literally just drew if you could properly detect the shapes and how they were nested within each other.

It's really hard to do detection with hand drawing. We might make special stickers for common UI elements (botton, tab bar ...). The color coding suggestion is great, thanks!