I’ve been experimenting with image generation with ChatGPT Plus plan in the web UI. The first image I make is exceptional quality, crisp, detailed and looks great. The problem occurs when I want to tweak something in the image. It can be a tiny thing, and usually it can adjust it to my liking, but the rest of the image suffers a severe quality degradation. The effect is even more pronounced if I again edit the image (that was already edited once). The more times you edit an already edited image, the worse the quality becomes. It loses the crisp details, the image gets more blurry, and what seems to be compression artifacts start appearing. The effect is very noticeable with just 2 edits to the same image. This has led me to breaking out Photoshop and making the time-consuming edits by hand using the original crisp generation because, even though the edits are usually correct, the drop in quality for the rest of the image is unusable.
Hi and welcome back @04894fd239f8419198ca,
That’s actually something I’ve noticed too: the more times I edit the same image, the more the quality tends to degrade, especially when editing through ChatGPT. I haven’t noticed the same degree of quality loss when using the Image API for edits, though.
You can read more here if you’re interested: Create image edit | OpenAI API Reference
If you want to continue editing through ChatGPT, I’d recommend returning to the original image for each new attempt rather than repeatedly editing the latest result. Try combining all the changes you want into one specific prompt.
You can also ask ChatGPT to write a consolidated editing prompt for you, review it first and then submit it together with the original image. That may reduce the number of editing passes and help preserve more of the original quality.
I’d also recommend checking this topic: Collection of GPT-image-generator 2.0 issues, bugs, and work-around tips (check first post) . You can add the issue you’re experiencing there too.