I'm curious if there's a future where ChatGPT could create images while keeping all the details intact instead of making random changes. I understand there's technologies like Stable Diffusion, but it seems that ChatGPT doesn't replicate images exactly due to some randomness in how it operates. Will there ever be a way to do this, or would we need a different kind of AI for that?
2 Answers
I think eventually we'll get closer to that, but it's not going to be as straightforward as we might imagine. AI has made progress from fully altering images to generating similar versions, but getting identical output every time isn't something we've cracked yet.
I can relate! I tried getting a ChatGPT-generated image of my late kitten, specifying that I didn't want any white on her chin, but it still turned out different. It often misses those small details, even when I provide clear instructions. It can be frustrating!
But why is that? Can't OpenAI just use the entire image for context and edit specific parts more precisely?