In a more perfect world AI would exist to make our lives better, e.g., using the capabilities of analyzing vast amounts of data and spotting patterns to develop cures and treatments for diseases that have plagued us for decades if not centuries. In the real world however, it exists to make money, in a few words, because of greed. PCs weren’t invented so you could play games and such at home – their original purpose was to save biz money and so increase profits. Same thing with AI. AI will allow biz to eliminate more jobs than the widespread adoption of PCs years ago, just as it will be used for more negative purposes, like crimes including fraud on a much larger scale.
But like so much else, AI is just a tool – it’s the people using it who are good or evil. And it’s a tool that’s too useful to ever be abandoned. The genie’s out of the bottle, so might as well use it when you can – frankly, whether you like it or not personally isn’t going to matter – so here’s hoping you use it for good.
One of the ways that you can use AI is in photo / image editing & creation. Today AI can create images, music, video, programming code, and writing etc., all of which are dual use, can be good or bad, so your personal ethics, and the boundaries you set, are more important than ever. Companies like Topaz and DVDFab [which was generous enough to give us a year’s license for their AI app] offer mainly noise removal and overall image enhancement, while plugins using the Stable Diffusion and DALL-E image generators are available for Photoshop, the GIMP, Krita, and soon probably other apps. Photoshop and Luminar Neo take the lead however, letting you do things you never imagined editing & restoring photos. Neither is cheap -- you **might** find a site that still has an old offer for Luminar AI for free – but if you’re serious about restoring old photos, you pretty much don’t have a choice, they’re that good.
The photos I’m working on are film prints, and like most [all?] prints, the amount of sharpness and detail is severely limited by film grain. Photoshop now has 3 tools to remove noise: in Lightroom & Photoshop Camera Raw there’s an AI denoise that works like the Topaz AI Denoise app & plugin; Photoshop’s Neural Filters include a Photo Restoration [Beta] plugin that seems to do roughly 8-10 times as much grain removal; and Lightroom / Camera Raw has a manual denoise that has a great smoothing effect after using the other two. After using the Photo Restore plugin followed by Luminar Neo their sharpening effects can leave areas of the photo somewhat mottled, e.g., the shading on bare arms, and that last denoise step, applied selectively using a mask, can work wonders on smoothing that out.
AI isn’t really intelligent, but it’s good at spotting patterns, so when it *sees* the basic outlines of a face for example, it looks in its database for similar patterns, and then makes a generally very good *guess* at what a higher resolution version of that face would look like. The AI in Photoshop and Neo can do this sort of thing for the entire photo with generally very good results, but they also make mistakes, so you must proof the results carefully. As above, the shading on bare skin often needs a little work, often with that 3rd denoise step, sometimes using the heal & remove tools. If the head is partially turned, Photoshop’s restoration AI can have problems with glasses, and both Photoshop and Neo can occasionally mess up ears.
Besides removing noise, and optionally scratches, the Photo Restore plugin does a remarkable job with faces, reconstructing detail that’s only hinted at in the original print. [Note: unfortunately, it does not like faces looking to the side.] Neo can also do neat things improving the lighting, whitening teeth, and even changing eye color. Lightroom and Photoshop’s Camera Raw filter [basically the same thing as Lightroom without the photo cataloging features] have presets to select body parts and do things like whiten teeth, but they’re a work in progress, & not as good as the features Neo offers.
Photoshop has long had a healing tool, and it can often help, but the new Remove tool works like that on steroids. Yes, you can paint over something, and it will remove that object [or person etc.] from the photo, but just click it in place once and it works like the healing tool on whatever’s under the cursor. The AI image generation in Photoshop Beta lets you replace the background after selecting it to exclude the people &/or objects you want to remain visible in the photo. I could see using it on shots where the window’s blown out, with nothing outside the window visible [assuming I knew what was outside that window – I’m restoring, Not creating a new composition], but I haven’t tried it yet.
Using the image generation tool to add or replace parts of the photo has blown me away! Working on a photo my mom took of our son, she used her camera’s flash, which overexposed his entire face. One result was that his right eyebrow was missing -- nothing there at all. So, I added one, and it matched (!). A bigger problem was his eyes, which were not only all red, with little to no detail, but a flock of hair partially obstructed his right eye. Especially with that flock of hair, Neo couldn’t do much with his eyes when I changed the color to blue. So I asked Photoshop to generate new eyes, and it did! Though I had specified blue, they were brown, but Neo promptly took care of that.
Working on another of mom’s photos, it showed 4 boys, our 2 and their cousins, with three of them under a partial shade, and one out in the full, Florida sunlight. The right edge of the photo was blown out [white], and with the boys off center, the 4th boy was almost touching that edge – the corner of the photo under his feet was completely blown out, a solid white. Since I had to create something to fill in that corner, I figured why not extend the whole photo, centering the boys? The image generation tool nicely did the trick, creating a very believable extension of the existing background.
A third photo my mom took was more challenging, with our son jumping into a ball pit that was surrounded by variably sagging netting. There was something or someone on her left when she took the picture, and because of the camera’s flash, on the left edge of the photo there was this big, white blob. I couldn’t crop [cut off] that part of the photo without also losing our son’s arm, and that sagging net made cloning, replacing the blob with portions of the rest of the photo, basically impossible. In this case I cropped the photo as much as I could while keeping our son’s arm, then extended the background, including the complete portion with the blob. It did a very believable job, including a new version of our son’s arm, though I then removed the new arm & did a cut & paste of the original.
The first photo would have been usable, with Photoshop’s AI simply making it better, while the 2nd would have been marginal if it were not for the AI. With the third example however, I managed to save a photo that was ruined otherwise. A much better artist than me *might* have been able to paint over that white blob, but I highly doubt the result would have been as believable.
These 3 photos also illustrate the scary part in all of this… Before AI, someone skilled / trained at this sort of thing could detect a photo that had been altered probably 99% of the time, maybe more – after or since AI, that might drop below 10%. To counter that the industry has been working toward consensus on a sort of signing protocol or standard, which would work something like the authentication process used for web sites and program files, e.g., .exe and .dll files. While not foolproof – most things can be hacked – it would give someone who cared enough to look a reasonable assurance that what they see in a photo is in fact genuine. There will be some amount of upheaval nonetheless, as I wonder what will happen in the UK with their [over?] reliance on CCTV. If anyone can very easily recreate footage, how hard would it be to substitute that for the original?
