Copy link to clipboard
Copied
In the last Beta versions Generative fill brings worse results:
It often does not make the new seamless to the existing.
For instance this very simple sky  – I want to replace the dark upper left.
Original:
Results:
 
 
Copy link to clipboard
Copied
Yep, I've noticed the same thing.
Copy link to clipboard
Copied
I have been getting the impression that a lot of people use Gen Fill when Content Aware Fill would be a better fit. At least we have some control of the process with CAF. [EDIT] ...and it does not chew through your Gen Credits.
Copy link to clipboard
Copied
Yeah because generative fill is a lot more convenient because content aware fill have much more demanding prerequisuites when it can be used. I don't know exactly what but my experience is that unless I am working with a single layer content aware fill is genenrally greyed out.
Copy link to clipboard
Copied
I heard a rumor that because gen.fill was trained on images without consent of owner, they had to roll back to previous version. My generative fill has been so ugly and no longer understands my prompts, cannot recognize pre-existing scale of objects, and ignores basic instructions. It's as rough as early 2023 attempts.
Copy link to clipboard
Copied
Same on here. The output since V25 is absolutely useless. The quality went from top-notch to garbage.
Copy link to clipboard
Copied
The situation has gotten worse with the latest non-beta version of OSX (26.0.0). Despite not selecting the option to remove older versions, all my previous versions were uninstalled. When I try to reinstall an older version, the generative fill feature automatically opens the newer one instead. It’s been quite a hassle.
The quality has significantly declined. This version generates far more nonsensical fills compared to the previous one, which is really disappointing.
Copy link to clipboard
Copied
And not only that, it's Overcensured now.