AI, or artificial intelligence, has been hitting the headlines over the past year. AI tools are everywhere, lurking in software on a computer near you. You could almost say it has become an obsession, yet another reason to face the apocalypse. Doom-mongers forecast that, left to its own devices, it will eventually wipe out humanity. So there.
We’ve covered AI over the months with stories on DALLE-2 and ChatGPT with the written word. Everyone is jumping on the bandwagon, including Amazon Web Services:
AWS Generative AI Innovation Center, a new program to help customers successfully build and deploy generative artificial intelligence (AI) solutions. AWS is investing $100 million in the program, which will connect AWS AI and machine learning (ML) experts with customers around the globe to help them envision, design, and launch new generative AI products, services, and processes
Amazon Press Release
Now Adobe is in on the new theme with updates to Photoshop and Lightroom Classic. I’ve been taking a look at what nifty business I can get up to with the beta version of Photoshop 2023.
This prototype version of Photoshop incorporates a feature known as Generative Fill, allowing users to insert AI-generated images into photographs.
AI Tools
As a photography enthusiast for decades, I have never been shy of innovation. Having used or tried most photographic innovations over the decades, I could not resist the opportunity to dabble with the new AI tools.
When I found new tricks in Adobe’s flagship PP products, I couldn’t resist. In LightRoom Classic, find the new options in the Develop module. There are no fewer than 23 adaptive options, such as ‘Whiten Teeth’ and ‘Storm Clouds’
These function in the same way as presets, although they differ from presets in that they offer the opportunity to control the intensity of the application of the adaptive change through slider control. So you can have differing levels of pearly whites and storm-laden clouds.
The two photos below show the application of the Lightroom adaptive module. First up is the original image taken in the local Bouddi National Park one stormy afternoon.
Enhancements
The second is the same photo but with the storm clouds slightly ‘enhanced’ using the adaptive function. The LR adaptive function does not artificially generate content. It can, for example, only ‘enhance’ existing storm clouds. You cannot turn a cloudless sky into a stormy sky at the click of a mouse. That is where the Photoshop generative fill comes in.
For me, using the Lightroom adaptive options is the same as using presets, and arguably you can achieve the same result. Admittedly, it cuts down the work compared to using the existing tools in Lightroom and Photoshop. So the Lightroom adaptive options get a ✅Pass from me.
Generative Fill
The Photoshop Generative Fill function is a different story. Currently, it is only in the beta version of Photoshop, but let me demonstrate it.
The image above shows the Skillion, a headland in Terrigal, New South Wales, Australia, where I live.
Old shipwrecks litter the coast around these parts. All of which are now underwater until I came along with my AI and raised at least one from the depths. As you see from the following image, there is now a large rusting hulk on the rocks below the Skillion.
Adding the shipwreck was straightforward. I used the lasso tool to define the area where I wanted the wreck. Then I typed ’shipwreck’ into the dialogue box and clicked on Generative Fill. Lo and behold, the wreck appeared and blended in. It certainly looks as if it is part of the scenery.
Encouraged by this, I asked Photoshop to insert a large cruise liner sailing past the Skillion. Again, the generated image blended in well, as you can see below. Convincing, eh?
The shipwreck and the cruise liner were the best of my AI fill attempts. In one less successful effort, I perched a lighthouse on the top of the Skillion — to help prevent those generative shipwrecks, you understand — but it looked artificial and unrealistic. I also tried putting a roller coaster up there but with pretty much the same result.
The Cloud handles this Generative Fill computing and occasionally stutters. I received a few notifications to the effect that the servers were very busy and I should try again later.
Red Card
This was a fun exercise, but I cannot see myself using Generative Fill for real for a very simple reason: the resulting photos are artificial. They are fakes. There is no other description. However, such techniques have their place in demonstrating a point, provided their use is made clear. Indeed, governments are sitting up and taking notice.
A proposed UK law could mean that all artificial intelligence (AI) generated photos and videos must be labelled. The Times reports that U.K Prime Minister Rishi Sunak is currently considering the legislation in a bid to regulate the fast-emerging technology and combat the threat of deep fakes
So it’s a ❌Red Card for Generative Fill from me.
I should emphasise that I was only playing with these new AI tools simply out of interest. There will be many Macfilos readers who are probably proficient and who don’t share my caveats on their effectiveness.
This short story came about only because editor Mike saw my shipwreck photo and suggested I write a piece on my experience. This is certainly not a fully reasoned review of AI tools in photography.
What do you think? Do you see a place for AI tools in post-processing, or do you think this is a step too far?
See the 100+ articles on Macfilos by the same author
Visit John Shingleton’s blog, The Rolling Road
A cup of coffee works wonders in supporting Macfilos
Did you know that Macfilos is run by a dedicated team of volunteers? We rely on donations to help pay our running costs. And even the cost of a cup of coffee will do wonders for our energy levels.
Dear John, thank you very much for sharing your experiences. In a way, this is threatening and promising at the same time. We can soon save to much time in post-processing and lose so much credibility for our results. We will see much more of this within months rather than years. BTW, have you tried the new AI base noise reduction feature of Lightroom? This one is impressive to say the least. Makes older cameras such as a Leica M (240, 262) all the more attractive. Thanks again, best wishes, and it’s great to read from you again. Jörg-Peter
I agree with the summaries so far
For years we’ve been trying to successfully remove objects (eg orange anoraks from glaciers and inside cathedrals) and now we put them in again!
Endless fun?
1. Now you see me, Now you don’t Now…
2.The game of Battleships takes a new turn
3.(now it’s your turn)
Can we get back to the real world of natural composition in the art of hobbyist and competition photography
I really don’t have much interest in creating an imaginary image of something that never was there and isn’t, but is a fantasy image. It is just the art of illusion.For some unknown reason I’ve never been wowed by magic acts either. If I made an image of a car floating over Buckingham Palace with AI it might get someone’s attention, a party trick, but beyond that? It all seems to be about..Hey! Look what my technology can do! Isn’t that cool! But why?
‘It’s been Photoshopped’ is already part of the common language.
Coin the neologism for AI ‘enhancements’ and win instant fame. But only for ten minutes.
At the moment, there appears to be what I would call “AI Assist” (my term), for lack of a better label. This allows making enlargements without increasing noise, perhaps make complex masks much more easily, improving the rendition of faces upon re-sizing, and so on. These are labeled “AI” by their developers, but are akin to sophisticated darkroom techniques in that they do not change the basic ingredients of the image.
Then there is PS’s clone brush. This is not called AI (but I could image a more sophisticated version that would utilize AI). I use this frequently to remove, extend, etc. objects in a photo. This is done to enhance the image, and I always mention this if the image is being published.
OTOH there are plug-ins that I occasionally use, such as one which substitutes clouds for a blank sky. I find that morally a bit troubling, although I may have tried to do this way back in the wet darkroom days. In any case, I would mention this, depending on the use of that particular photo. NB: these masks do not always fit or look realistic.
As for using AI programs which put something into a photo which was never there, and look convincing, then we are crossing a Rubicon, IMO. Such usage definitively needs to be mentioned. Is this a step too far? It would depend on how the image is being used, I guess: used for fantasy and labeled as such, versus used falsely for a document. Fact or fiction? Think of the Soviets crudely removing a person from a photograph at their May Day Parade atop Lenin’s tomb because they are now a non-person. Only this is worse, because it won’t be obvious.
The most basic question is: what is a photograph? We used to say it boils down to the negative. With digital, we could no longer say that very easily. With intrusive AI, things are further unmoored from reality, unless all the digital steps in making that image are saved. That is totally unrealistic. Such an image could never be used in a legal case, and ought not to be usable in any public venue unless labeled.
Thank you John, for so aptly presenting what we will be facing.
Ed
.
I used to use a little standalone program called ‘Pearly Whites’, around 2011 or earlier, from a firm called Image Trends Inc. It was a Photoshop Elements plug-in, and would find teeth – in any photo which showed them – and could whiten them to whatever degree you specified, as – back then – digital pics tended to make teeth look somewhat yellow-ish.
Image Trends made another program which ‘un-curved’ extreme wide-angle or fisheye shots to make everything look straight again (..although rather edge-stretched).
I haven’t used either of those programs for years ..th-thanks for the memories!