The Image Generator is... well lobotomized ?
retained text
I'd like to share something unheard of. Since October 28th, I haven't been able to work with ChatGPT—I've been using it since the second generation of DALL-E, and let me tell you, NEVER has the generator worked this poorly... I can't get it to do anything except for extremely generic visuals... it understands nothing.... It doesn't take reference images seriously, and workflows I built over months that used to run reliably don't work anymore either... and this is not mere beginner promptworkflow.. trust me.. It literally feels like it's running behind the walls.. on flash, lite, or turbo models.. as if the generator itself was literally downgraded to smaller steps....and cutted of from datasets.... Every now and then it manages to produce something that slightly works, as if remnants of the old dataset kick in.. but other than that... disastrously ridiculous, I'd even say tacky results...even if i change the prompt the writing .... nothing.. same ugly generic style and little bit make up on the surface from the refrence... I don't understand these moves by OpenAI.. where this is heading and why.... I don't know if anyone else is experiencing this like I am.... but I'm sick of it, support is ignoring me... and when they do send something, it's just a bot, so it tells you nothing... I don't get the direction of this.. people who only do photorealism shouldn't comment, this isn't about generic styles.
document fetched 2026-11-08 00:19 · canonical https://reddit.com/r/ChatGPT/comments/1xpylia/the_image_generator_is_well_lobotomized