Photoshop Beta Generative Fill AI

[JimH123]

Going back to the squirrel shot, I can see a quite obvious line down the image where something has changed. The backgrounds are a little different and clearly delineated. It was impressive till I saw that . The lighthouse image looks weird. Superimposed I'd say. But given you spent so little time but still got a half decent return, it bodes well for all sorts of shenanigans down the line. As long as no one gets hurt!

Not sure what you are seeing on the squirrel image. Looks fine to me. When you delete something by drawing a selection around it, Generative Fill carefully works to match the area around. But there is a weakness today and that is it can oly fill 1024 x 1024 pixels, Perhaps the difference is resolution is what you are seeing.

 

[fuji_phil_e]

Going back to the squirrel shot, I can see a quite obvious line down the image where something has changed. The backgrounds are a little different and clearly delineated. It was impressive till I saw that . The lighthouse image looks weird. Superimposed I'd say. But given you spent so little time but still got a half decent return, it bodes well for all sorts of shenanigans down the line. As long as no one gets hurt!

Not sure what you are seeing on the squirrel image. Looks fine to me. When you delete something by drawing a selection around it, Generative Fill carefully works to match the area around. But there is a weakness today and that is it can oly fill 1024 x 1024 pixels, Perhaps the difference is resolution is what you are seeing.

Whatever the theory you're putting across, it's quite clear to me:

 

[JimH123]

1. Antique Eruption wrote:

Going back to the squirrel shot, I can see a quite obvious line down the image where something has changed. The backgrounds are a little different and clearly delineated. It was impressive till I saw that . The lighthouse image looks weird. Superimposed I'd say. But given you spent so little time but still got a half decent return, it bodes well for all sorts of shenanigans down the line. As long as no one gets hurt!

Not sure what you are seeing on the squirrel image. Looks fine to me. When you delete something by drawing a selection around it, Generative Fill carefully works to match the area around. But there is a weakness today and that is it can oly fill 1024 x 1024 pixels, Perhaps the difference is resolution is what you are seeing.

Whatever the theory you're putting across, it's quite clear to me:

I see it now. I wasn't looking at high magnification. Gen Fill is not duplicating the noise in the image.

 

[Guy Dagar]

Going back to the squirrel shot, I can see a quite obvious line down the image where something has changed. The backgrounds are a little different and clearly delineated. It was impressive till I saw that . The lighthouse image looks weird. Superimposed I'd say. But given you spent so little time but still got a half decent return, it bodes well for all sorts of shenanigans down the line. As long as no one gets hurt!

Not sure what you are seeing on the squirrel image. Looks fine to me. When you delete something by drawing a selection around it, Generative Fill carefully works to match the area around. But there is a weakness today and that is it can oly fill 1024 x 1024 pixels, Perhaps the difference is resolution is what you are seeing.

Whatever the theory you're putting across, it's quite clear to me:

Yep, the process has some flaws that the photographer has to deal with. Taking a closer look, it seems to be an issue of resolution and/or noise.







.

 

[JimH123]

Going back to the squirrel shot, I can see a quite obvious line down the image where something has changed. The backgrounds are a little different and clearly delineated. It was impressive till I saw that . The lighthouse image looks weird. Superimposed I'd say. But given you spent so little time but still got a half decent return, it bodes well for all sorts of shenanigans down the line. As long as no one gets hurt!

Not sure what you are seeing on the squirrel image. Looks fine to me. When you delete something by drawing a selection around it, Generative Fill carefully works to match the area around. But there is a weakness today and that is it can oly fill 1024 x 1024 pixels, Perhaps the difference is resolution is what you are seeing.

Whatever the theory you're putting across, it's quite clear to me:

Yep, the process has some flaws that the photographer has to deal with. Taking a closer look, it seems to be an issue of resolution and/or noise.



.

Might not be a probem if the image had been properly denoised before doing the removal.

As an experiment, I downloaded your original as a JPEG and used PhotoAI on it, and then removed the stuff on the left. It was harder to tell where the line was. But had this been done with a RAW file, it would have been a lot better.

Noise is better being removed from the RAW file. Once it is a JPEG, noise is sort of baked in.

 

[Guy Dagar]

JimH123 wrote

Might not be a probem if the image had been properly denoised before doing the removal.

I'm thinking that also, but I used DxO PureRAW which generally gives good results. Looking at that edit, noise seems to be mitigated "properly."

 

[JimH123]

JimH123 wrote

Might not be a probem if the image had been properly denoised before doing the removal.

I'm thinking that also, but I used DxO PureRAW which generally gives good results. Looking at that edit, noise seems to be mitigated "properly."

What noise option did you use for DxO PureRAW? It should be capable of getting rid of it. But there certainly is a lot of it showing in the one you supplied.

Are you sure you didn't just use the original RAW for the Generative Fill attempt?

 

[fuji_phil_e]

It's really interesting stuff. As has been said re. the squirrel, raw denoising would have probably prevented the issue. It's only a matter of work flow, so can be easily fixed in future efforts.

 

[Guy Dagar]

It's really interesting stuff. As has been said re. the squirrel, raw denoising would have probably prevented the issue. It's only a matter of work flow, so can be easily fixed in future efforts.

Raw denoising was applied as far as possible so perhaps the generative fill is so noise-free that it shows a difference. However, the original squirrel photo was at ISO 20000, and the jpeg is heavily cropped. I chose it because of the challenge to create something from a photo that would otherwise be mundane. Here's a copy of the original after denoising, before cropping. Yes, if this had been a photo taken at perhaps ISO 6400 and not cropped or at least zoomed in with a longer lens, the denoising would have been more successful. You can only denoise an image so far before it becomes very muddy. And I'm not sure there is any particular formula for "proper" denoising, you just have to try to balance the noise with the sharpness.

 

[PieterB]

My sister is a complete digital illiterate and she said that she found it rather stupid that a seagull was sitting on her son's head. She didn't believe the birds were fake.



Than I upped the pictures a bit and made a really completely stupid picture of my brother and his wife. Of course, she didn't watch it carefully but eventually she realised that picture was completely over the top and was probably fake.



--
No life without a camera.

 

[Mark B.]

My sister is a complete digital illiterate and she said that she found it rather stupid that a seagull was sitting on her son's head. She didn't believe the birds were fake.



Than I upped the pictures a bit and made a really completely stupid picture of my brother and his wife. Of course, she didn't watch it carefully but eventually she realised that picture was completely over the top and was probably fake.

LOL! I actually don't mind pics like this that are over-the-top and amusing. I especially like the squirrels running up his legs.

 

[Guy Dagar]

My sister is a complete digital illiterate and she said that she found it rather stupid that a seagull was sitting on her son's head. She didn't believe the birds were fake.



Than I upped the pictures a bit and made a really completely stupid picture of my brother and his wife. Of course, she didn't watch it carefully but eventually she realised that picture was completely over the top and was probably fake.

LOL! I actually don't mind pics like this that are over-the-top and amusing. I especially like the squirrels running up his legs.

I agree! His intention wasn't to fool anyone, but to make photos that are humorous and absurd. Nice work! Very creative. Art.

 

[PieterB]

When I sent her this pictures, she seemed to realise that this couldn't be true. Even the glasses with whiskey and beer were just mugs of tea and coffee and the baby of course had a baby beaker.

The things that were added were.

• The knife on the wall.
• The pink wall decoration.
• The hats
• The cigars.
• The necklace and earrings.
• The dog
• The baby on the right.
• The dolls.
• All the glasses.
• The jar of booze.

--
No life without a camera.

 

[Mark B.]

When I sent her this pictures, she seemed to realise that this couldn't be true. Even the glasses with whiskey and beer were just mugs of tea and coffee and the baby of course had a baby beaker.

The things that were added were.

• The knife on the wall.
• The pink wall decoration.
• The hats
• The cigars.
• The necklace and earrings.
• The dog
• The baby on the right.
• The dolls.
• All the glasses.
• The jar of booze.

Love it!

There's a guy on a FB page for a local shore town that's well-known to regulars for adding elements that don't belong. It's hilarious when people take it seriously; one of my favorites was when he superimposed kangaroos hopping around on the beach.

 

[ganthercage]

If you know me, you know I like controversial subjects, and this is one of those: the new Photoshop Beta Generative Fill feature. I've been getting notices about Photoshop Beta for some time and have been ignoring them. Today I decided to check it out. After installing it, and figuring out how to access Generative Fill, I gave it a try. The first photo is the original, which is to my way of thinking pretty mundane and I'm not sure why it was still in my catalog. The second one is my first attempt at Generative Fill. It's impressive, without even trying—it only took a few minutes to render the second photo; I used Generative Fill to add the lighthouse, and Sky Replacement to add a sky from my sky collection.

Here's my opinion of it, and of AI in general as applied to photography. I think in some respects it can be very useful to improve a photo that has a good basis, one that just needs a little something extra. I have used Photoshop's Sky Replacement feature before utilizing my own sky photos; it's not an easy process and often not successful. At other times, I've used double exposures to blend two or more of my photos to make a unique image, something I did a lot of back in the film days in the darkroom and with a slide-copier attachment on a bellows. Generative Fill can also be used quite successfully to remove an object from a photo and fill in the blank, a very useful tool.

But Generative Fill goes way beyond that—it can add all sorts of objects and subjects to a photo that come from somewhere else, making it something other than a real photo that you or I actually took with our camera(s). It worries me that the true art and skill of image-making is going beyond the skills and artistic talents of a photographer. It's worse than the advent of cell-phone photography. I know there are some AI apps online that create images entirely without more than a little input from a user, certainly a big money-making cost-saving step forward for ad agencies and stock photo houses and anyone that buys photography, and maybe there will be new agencies that create and sell AI photos.

Please, don't get me wrong, I'm not suggesting that the photo I used the Generative Fill on is anything other than an experiment to see what it can do. I know the image leaves a lot to be desired and it isn't going to win any awards. Just a test.

The topic of AI-powered image manipulation and creation is indeed a controversial and thought-provoking subject, and opinions on this matter can vary widely among photographers, artists, and the general public.

 

[PieterB]

If you know me, you know I like controversial subjects, and this is one of those: the new Photoshop Beta Generative Fill feature. I've been getting notices about Photoshop Beta for some time and have been ignoring them. Today I decided to check it out. After installing it, and figuring out how to access Generative Fill, I gave it a try. The first photo is the original, which is to my way of thinking pretty mundane and I'm not sure why it was still in my catalog. The second one is my first attempt at Generative Fill. It's impressive, without even trying—it only took a few minutes to render the second photo; I used Generative Fill to add the lighthouse, and Sky Replacement to add a sky from my sky collection.

Here's my opinion of it, and of AI in general as applied to photography. I think in some respects it can be very useful to improve a photo that has a good basis, one that just needs a little something extra. I have used Photoshop's Sky Replacement feature before utilizing my own sky photos; it's not an easy process and often not successful. At other times, I've used double exposures to blend two or more of my photos to make a unique image, something I did a lot of back in the film days in the darkroom and with a slide-copier attachment on a bellows. Generative Fill can also be used quite successfully to remove an object from a photo and fill in the blank, a very useful tool.

But Generative Fill goes way beyond that—it can add all sorts of objects and subjects to a photo that come from somewhere else, making it something other than a real photo that you or I actually took with our camera(s). It worries me that the true art and skill of image-making is going beyond the skills and artistic talents of a photographer. It's worse than the advent of cell-phone photography. I know there are some AI apps online that create images entirely without more than a little input from a user, certainly a big money-making cost-saving step forward for ad agencies and stock photo houses and anyone that buys photography, and maybe there will be new agencies that create and sell AI photos.

Please, don't get me wrong, I'm not suggesting that the photo I used the Generative Fill on is anything other than an experiment to see what it can do. I know the image leaves a lot to be desired and it isn't going to win any awards. Just a test.

The topic of AI-powered image manipulation and creation is indeed a controversial and thought-provoking subject, and opinions on this matter can vary widely among photographers, artists, and the general public.

Back in the day of pure analogue photography, pro photographers were also busy masking stuff and busy with little brushes making ugly stuff away. Not the amateurs of course but the professionals. The Sovjets were famous for it.

Look at this site.

https://rarehistoricalphotos.com/stalin-photo-manipulation-1922-1953/