I can see the potential for Generative Fill to make possible many other ways to fix problems
Jun 24, 2023 21:34:29 #
The following image is an IR image and for this one, it suffers from a problem called "hot spot". Some lenses have this problem, and others don't. It is not that easy to fix a hot spot and can be tried with a radial gradient brush in Lightroom where you attempt to place this radial gradient in exactly the right spot and to apply just the right amount of adjustments to seamlessly take out that hot spots such that it is no longer visible.
The second image is an attempt to do this using Generative Fill in which I drew a circle around the hot spot and let generative fill figure out what to do. It is sort of working, hot spot is reduced, but it is also subject to changing the shore line. Perhaps in time, there could be a keyword to fix a hot spot.
That also opens up possibilities for other fixes that it could be taught to do. Some possibilities might include fixing CA, or to fix astigmatism/coma distortion, and possibly other things too. A list could be made.
The second image is an attempt to do this using Generative Fill in which I drew a circle around the hot spot and let generative fill figure out what to do. It is sort of working, hot spot is reduced, but it is also subject to changing the shore line. Perhaps in time, there could be a keyword to fix a hot spot.
That also opens up possibilities for other fixes that it could be taught to do. Some possibilities might include fixing CA, or to fix astigmatism/coma distortion, and possibly other things too. A list could be made.
Original with hot spot
(Download)
A fix with Generative Fill that removes most of the hot spot
(Download)
Jun 24, 2023 22:43:26 #
In attempting the Lightroom Radial Gradient method, I think it turned out better than force fitting Generative AI to make it work.
Jun 24, 2023 23:24:12 #
I think the possibilities for this technology are almost endless, but this current beta version is just a toy. Low resolution is no good for any serious work, and of course it generates weirdness more often than anything useful. You only have to look at how far Midjourney has come to see the possibilities, but for now, I can't get too worked up over it. It is fun to play with on low res images, though.
Jun 25, 2023 00:25:04 #
terryMc wrote:
I think the possibilities for this technology are almost endless, but this current beta version is just a toy. Low resolution is no good for any serious work, and of course it generates weirdness more often than anything useful. You only have to look at how far Midjourney has come to see the possibilities, but for now, I can't get too worked up over it. It is fun to play with on low res images, though.
Working with small increments can work around the low res problem. But it would be nice to see some guidelines for picking keywords to make it do things as all I can do now is guess what might work. And often the keyword did nothing, and when I try the same thing without a keyword, it behaves the same (sometimes).
It will be interesting to see where this goes next. But you are correct. This is currently a toy. And as a new toy, I just have to try things out with it, but no plans for anything serious. And I agree, I get a lot of weirdness. Some results look good. Some results look like cartoons. And some results -- how can I even describe it?
Took my lake example and tried to find keywords that actually worked, Settled on an iceberg in the center of the lake with a polar bear on it. Hand sketched the tip of the iceberg, and it followed it. Hand sketched a polar bear facing the left, and it followed that too. At least silly things seem to work. Adobe has a ways to go on this.
Jun 30, 2023 08:44:01 #
Great AI vs gradient exercise. Thanks for sharing it. Love that you captured the polar bear on glacial float ice.😊
If you want to reply, then register here. Registration is free and your account is created instantly, so you can post right away.