I know this is a bit out of scope for these image editing models but I always try this experiment [1] of drawing a "random" triangle and then doing some geometric construction and they mess up in very funny ways. These models can't "see" very well. I think [2] is still very relevant.
No comments yet.