I’ve requested it before & will do so again. It would help a lot to be able to manually paint an overlay on non-identified faces to have them be treated with the Recover Faces models.
I realize that doesn’t guarantee magic. But it would be important to be able to give it a go. It’s frustrating when no face ID option is offered.
After I installed this patch, the Recover model no longer works like before. I have a 4090 GPU and was able to run Recover in preview for an entire large pic (over 3,000px/squared) in about 20-30 seconds max. Now preview doesn’t work (it juts hangs halfway through) and when I try to actually run the job which used to take 20 seconds or so, it now runs my GPU at 100% indefinitely. I’ve let it go at least 1o minutes. My GPU doesn’t ever go to 100% even with Video AI upscaling to 4K and 60 fps.
All other models including Redefine seem to be working fine. I have tried Recover while Face Recovery was off as well as on, makes no difference. I was able to get it to run once but I had to shrink the output to less than 500px/sq. which obviusly defeats the purpose of Gigapixel AI lol
Not that I know anything about painting, nor can I paint, but vague memories (from a regular school, not an art school!) of types of artistic style suggest that it could be so-called naïve art. The first picture in particular (perhaps the second one too, and many previous ones) without a landscape perspective looks so naïve. The creation of artificial intelligence using a certain given inspiration. AI probably wouldn’t have invented the Mona Lisa, but it can sometimes be suitable for illustrations of e.g. books, advertisements, etc. However, the result can often be unexpected. Well, even T. G. is no longer the certainty it once was…
The first image is a 2022 AI render, which in that app was purposely abstract in its results. I am always amazed at how Redefine can make these into something similar yet completely different. It’s fun to explore the GPAI renders to see what clever bits appear.
I wanted to learn about naïve painting and found (Wikipedia) a prominent representative, Henri Rousseau. I tried applying maximum Redefine to his self-portrait (1890) to see what it would do. I was really surprised (without any ulterior motive!). So, the rather hairy Topaz Henrietta . I have the impression that (maybe not only) high levels of Redefine go too often like a spinning wheel. Let’s say skillful paintings, but limited ingenuity. It would be nice to improve the Redefine function (but surely there are much more pressing tasks). Maybe better use of the text prompt would help?
Me too. But what about some other Rousseau painting? Like… The Dream. The girl (original, on the left) was his Polish lover (according to Wikipedia). Redefine-Max prefers a somewhat edited work, and what about those plants! I would say that the name The Dream is now more justified with them. (I admit, a little prompt helped.) I wanted to try what happens when the original already has felines and lots of plants (which Redefine likes very often – when it doesn’t know what to put where, it floods an image with flowers and adds some lions and tigers). So which painting is better?