This would be easy to patch the models to fix. Just gather a small amount of training data for these cases, eg. "change the clock hands to 5:30" with the corresponding edit.
replyThree tuple: (original image, text edit instruction, final image).
Easy to patch for editing models, anyway. Maybe not text to image models.
waste of money and effort, imo. there are +- infinity of such small things to fix.
replyIt's a common enough use case that it'll be added at some point.
replyIt probably comes up more than you think. Storyboarding, product placement, model images, etc.
It's not critical in the short term, but it'll wind up on their backlog for sure.