https://mastodon.social/@UP8/110607460518784045
when I put the corrected one on the wall together with a lot of prints it was not properly centered vertically (had to center the painting + shadow not just the painting) so I had generative fill draw another row of bricks at the bottom. I’m sure they don’t look exactly like the the corresponding bricks on the real wall but they are good enough for wall art
It adds up to another reason to keep my Creative Cloud subscription instead of looking for an alternative.
The thing is if you want to have it add a spaceship or a pretty girl or something like that to the photo it has to make more background to go with whatever you tell it it draw so it automatically draws something consistent with the scene without a prompt.
There is a whole research area about tools that do "editing" so you could tell it to do something specific to an image, say "increase the exposure by 0.2 stops" (not that you need a.i. for that but somebody might rather do that than find the function in the UI) or "add a fourth motorcycle to the three motorcycles that are parked in a row" and that's even being done for 3-d models with NeRFs and point clouds and such but that's not what generative fill does right now.
I mostly get nice fill results as well, but they are super aggressive about removing "unsafe results", even when I don't give it a prompt? I'm not filling in nudity or anything, so it's not something I'd pay extra for in this state, I don't need an AI nanny.
Mine's only a 1060 though, so it does take minutes for a generation, whereas the newer cards can do it an order of magnitude faster.
Of course, I wouldn't mind it not using credits if I could use it locally either. Hopefully they continue to improve SD's IP Adapter type stuff so it's as good as Photoshop in that regard.
In its current state, at least, Photoshop's generative AI requires a lot of iterations to tweak the image into what I need (due to them requiring technical accuracy). Charging credits for this would make it nonviable.