Luminous Landscape Forum
Raw & Post Processing, Printing => Digital Image Processing => Topic started by: Benny Profane on May 24, 2023, 01:54:00 pm
-
Photoshop introduces an AI function in a beta version of the new release that will make rank amateurs push button pro retouchers. Generative Fill is a major advance. Here's one video that demonstrates it's power https://www.youtube.com/live/63xd7qu83FU?feature=share Not really a Scott Kelby fan, but, he talks to the advanced amateur or aspiring Photoshop retoucher well. The image extensions out of thin air are just amazing. You can argue about the ethics or morals of all this, but, accept it, because it will be here for a long time. Don't fight the tech. One thing I'm still not sure of is the quality of resolution of the added image data.Does it match a nice 26mp Raw base image?
Now I really have to buy a new computer, so I can use this latest version of Photoshop. Looks like a new Studio for me.
A nice explanation in text: https://gregbenzphotography.com/photography-tips/photoshop-generative-fill-text-to-image-ai/
-
Does it match a nice 26mp Raw base image?
In this version, the AI is limited to 1kx1k resolution. Smaller selections can help but that's the current limit per selection prior to Generative Fill/AI.
-
In this version, the AI is limited to 1kx1k resolution. Smaller selections can help but that's the current limit per selection prior to Generative Fill/AI.
If the Generative Fill is being used to remove an object in a high resolution file as opposed to adding a brand new object, would it not use/sample pixels from the high res file just like using the Spot Healing Brush?
-
If the Generative Fill is being used to remove an object in a high resolution file as opposed to adding a brand new object, would it not use/sample pixels from the high res file just like using the Spot Healing Brush?
No.
-
I played with this new feature, starting with a scenic I generated using the free text-to-image DiffusionBee app for Mac. Incredible stuff!
As Andrew noted, there is a resolution change visible along the edges of the original in the final render.
Benny, these were done on a base model 2020 M1 MacBook Pro so your computer doesn't have to be THAT new ;-)
-
Another totally computer-generated example. More fun in DiffusionBee, this time extending the canvas using GF and adding GF elements in the PS beta, followed by adding a cinema glow in ON1 Photo RAW (https://on1.sjv.io/BXY0mB).
-
Sheep May Safely Graze
Attached.
-
Your excitement about the new Generative Fill feature in Photoshop is contagious! I completely agree - the power of this tool to extend images is just mind-blowing. I'm sure it will be a game-changer for both amateur and professional retouchers alike.
As for the resolution of the added image data, @Digitaldog mentioned that it's currently limited to 1kx1k per selection, but I'm hopeful that Adobe will continue to refine this in future versions. The tech world is moving so quickly, and we're constantly being surprised by the new advancements.
-
The 1Kx1K limitation may change in the future. Much has to do with the demand on the Adobe servers. And all this is 'beta' anyway.
-
No.
Too bad. I was comparing it with some of the other tools in Photoshop for removing "extraneous objects" and it can work wonders.
-
More of my own examples.
The Torii Gate was created in Photoshop alone, starting with a blank canvas, while my other examples started with the DiffusionBee results.
-
Another set showing GF for expansion as well as changing/adding elements.
-
Too bad. I was comparing it with some of the other tools in Photoshop for removing "extraneous objects" and it can work wonders.
It can, with a pixel-per-selection limitation at this time.
-
Or you can do this online with Adobe Firefly (https://firefly.adobe.com/), with commercial limitations.
I've been to this place ;-)
-
Back to PS only (with white canvas), GF'd result GF'd again to enlarge further and post-processed with Radiant Photo (https://radiantimaging.pxf.io/ORD6MQ) for color.
-
Or you can do this online with Adobe Firefly (https://firefly.adobe.com/) . . .
Adobe Firefly created my steampunk camera:
(https://live.staticflickr.com/65535/52892815146_7cb0021a81_b.jpg).
-
Very cool Chris! But it takes film, right? ;-)
-
Very cool Chris! But it takes film, right?
Collodion plates.
-
Collodion plates.
Of course!
-
Using the Realistic Vision V2.0 model (https://civitai.com/models/4201/realistic-vision-v20) in DiffusionBee and GF'ing an expanded canvas in PS (!!).
-
So going forward I guess we can say, "Don't take it, make it!" ;-)
-
Is Adobe eating itself? Possible. As an ex retoucher, I can see a lot of jobs eliminated, although offshoring to the third world via the internet has accomplished a lot of that.
https://archive.vn/2023.07.25-090929/https://www.businessinsider.com/adobe-ai-firefly-kill-graphic-designer-jobs-cut-seat-sales-2023-7
-
Is Adobe eating itself?
There are no right answers to wrong questions -Ursula K. Le Guin
-
Is Adobe eating itself? Possible. As an ex retoucher, I can see a lot of jobs eliminated, although offshoring to the third world via the internet has accomplished a lot of that.
https://archive.vn/2023.07.25-090929/https://www.businessinsider.com/adobe-ai-firefly-kill-graphic-designer-jobs-cut-seat-sales-2023-7
AI is not going to replace graphic designers altogether, but it is likely to change the way that they work. Graphic designers will still need to be creative and have a strong understanding of design principles, but they will also need to be able to work with AI tools.
-
I have been fooling around with the generative fill tool in the new beta version of Photoshop and I am finding it to have great potential. Right now it is working well at removing small items, such as trash cans, hoses, rubbish piles, polls, etc. from the image better than I could and in a matter of seconds too. I have even had success at removing whole cars so long as the car was in front of something with out much unique detail, such as a pavement or wall.
I am also find it is really good at expanding landscape images into panoramas, albeit with limited resolution at the moment. This could help with my backdrop business.
Expanding urban and suburban images though, there is still a long way to go. Likewise, AI really only can work at generating objects, believably, it has ample source material for and if it does not, the failure becomes obvious. For example, in the historic sections of Philadelphia (Dating back to the 1700s), many of the buildings use what is called a flemish brick bond. Although the generative fill can reproduce an American bond easily, perhaps even an English bond, it is really poor at Flemish (and Monk) bonds when removing objects in front of the wall. This makes sense since Flemish bonds went out of style in the 1800s and there are few images available in stock catalogs to train the AI on creating them.
Another, yet surprising, disappointment is generative fill is not great at removing electrical wires in front of buildings, especially if you are looking at the building on an angle. In my experiments it either messes up the siding or the parallax or both.
Of course this is a beta version, and I can only expect it to get better once the actual version is released.
With that said, we do need to get past the copyright issues first with either new laws or at least a few dozen court cases for working precedents to become established. Remember, AI, legally, can not create a new unique piece of intellectual property. Only humans can do that. So, by law, any AI produced work is going to be an infringement against someone's (or many's) copyright.
Although Adobe appears to be taking the high road by only allowing their AI to be trained using images from their stock library, others are not so on the up and up, like Stable Diffusion.
Getty Images sues AI art generator Stable Diffusion in the US for copyright infringement (https://www.theverge.com/2023/2/6/23587393/ai-art-copyright-lawsuit-getty-images-stable-diffusion)
That lawsuit is for $1.8 trillion btw.
For the amateur, this is probably not a big deal. For myself, and other professionals, who will own the copyright of an image that utilizes AI is a big deal that no one really knows the answer to yet. I realize Adobe going about this in such a way that it will remove the liability of a copyright lawsuit from using an "AI image," but questions around if/how those images can be registered with the copyright office are not going to be answered for a number of years. With many of my clients, this is not that much of an issue. Hollywood though is a cut throat industry, so for my backdrops I need the protection registration offers.
-
It's now out of beta! But it sounds like it's a metered experience (?).
https://blog.adobe.com/en/publish/2023/09/13/ai-creative-cloud-release-pricing-update?sdid=VG52KCB7&mv=social&mv2=paid-owned
My first render in the new release, from "babbling brook in the fall woods with shallow depth of field, soft lighting, log cabin in the distance"
-
It's now out of beta! But it sounds like it's a metered experience (?).
Yes and still limited to 1K x1K.
-
Yes and still limited to 1K x1K.
Thanks Andrew, though it's really easy to keep expanding (this example takes my original to 4033 wide @ 300ppi, reduced here for posting).
Also, I'm stuck on a 2017 i5 Intel SSD iMac right now and the speed for GF in PS is quite good (backend processing, yes?). I did a local Diffusion Bee render earlier and it wanted 15 minutes!
Back home on my M2 Mac Mini it's less than a minute for DB...
-
I get around the size limitation of PS GF (and Diffusion Bee small renders) by upscaling in Photo AI (https://www.topazlabs.com/shop/ref/5/) (in this case, 4X from a PS GF done today).
-
The new Photoshop 2024 and Generative fill (text-to-image) from a blank canvas. A day at the AI beach!
-
Forget doing AI people, stick with octopi...
-
Just got Photoshop version 2024 and it needs apples system 13.4 to work well...
I guess it shows how important AI is getting that Abobe urges people to buy the computers that are not even on the market yet.
Ai is the most important change to the computerworld in a long time, making older computers unfit.
I suggest to wait till the M3 is around if you want to buy anyting from Apple that can handle AI properly.
-
Just got Photoshop version 2024 and it needs apples system 13.4 to work well...
I guess it shows how important AI is getting that Abobe urges people to buy the computers that are not even on the market yet.
Ai is the most important change to the computerworld in a long time, making older computers unfit.
I suggest to wait till the M3 is around if you want to buy anyting from Apple that can handle AI properly.
And now some data:
https://helpx.adobe.com/photoshop/system-requirements.html
-
and your provided data confirms what i said...
-
and your provided data confirms what i said...
Except for the bit about 13.4 (or the GPU needs for all users), but whatever. ;)
And zero issues with AI on a 2022 M1 MBP (YMMV).
-
I am not talking about issues,
It is remarkable that Adobe recommends a system as new as 13.4 ( 4 months old)
for sure all cards are on AI when developing the next M3 mac
-
Adobe and Apple (among others) recommend the latest OS but have shown the minimum requirements; so what? Nothing remarkable, and why the need for M3? M1 handles AI just fine for some of us!
-
Most of the Photoshop AI stuff is done in Adobe servers in the cloud, not locally, so no need for the latest and the greatest.
-
Adobe and Apple (among others) recommend the latest OS but have shown the minimum requirements; so what? Nothing remarkable, and why the need for M3? M1 handles AI just fine for some of us!
Remarkable is also that when i opened photoshop 2024 for the first time i immediately got an adobe pop-up that said: photoshop works best with 13.4 Ventura.
never had that before, ever. (I am on 12.6 Monterey)
-
https://helpx.adobe.com/photoshop/kb/photoshop-and-macos-ventura.html
-
The notification of the upcoming GF "subscription within a subscription" kind of took the fun out of using PS to do AI (for me), so I went back to Diffusion Bee and installed a few additional model sets, then banged this scenic out in minutes on an M2 Mini (16/500). All locally generated, no Adobe cloud, parental controls or subscription.
Of course PS makes it so easy to expand the canvas now so I had to do that as well... Then I added artsy as a frosting on the cake ;-)
-
Number of AI creations:
If you have the full Creative Cloud subscription, which gets you access to all of Adobe's software for $55 per month, you can produce up to 1,000 creations a month. If you have a single-app subscription, for example to use Photoshop or Premiere Pro at $21 per month, it's 500 creations a month. Subscriptions to Adobe Express, an all-purpose mobile app costing $10 per month, come with 250 uses of Firefly.
https://www.cnet.com/tech/computing/photoshops-firefly-generative-ai-arrives-with-a-creative-cloud-price-hike/
-
Thanks Andrew, that's doable ;-)
I have Colin Smith's video queued up but haven't watched it yet...
-
If the Generative Fill is being used to remove an object in a high resolution file as opposed to adding a brand new object, would it not use/sample pixels from the high res file just like using the Spot Healing Brush?
There seems to be a qualitative difference in how GF works. I've tried removing an object using Content Aware Fill, and then with Generative Fill. The latter tends to do a much better job.
It also seems, judging from my experiments, that when removing an unwanted object, you get better results if you make a fairly tight selection. I got very nice results by precisely selecting the object and then Select > Modify > Expand by just 5 or 6 pixels. Then run Generative Fill. Here's an example of how that worked: