Hacker News Reader

Hi Alex, great catch -> We didn't solve consistency, but we saw that if you regenerate a few times, you usually get something that's visually similar. Right now AI artists all do loads of postprocessing - using AI, so later we might have a "smear" feature that inpaints the inconsistent part. Let me know if you have thoughts on this

View on HN

People can generate a few good samples of what they want a character to look like and then interrogate clip to get a more detailed prompt that makes it more consistent. It increases the prompt sizes a lot, but I don't think there's an easy way to solve this.

Maybe you could build a UI that semi-automates this process?

Could eventually use GPT4vision to review the output of the art to see if all the panels are consistent?

Reply on news.ycombinator.com