It's absolutely mind-blowing how much artificial intelligence has come to play a central role in photography. There's the presence of it in everyday usage in smartphones, with things like portrait mode and night sight, but now, even more powerful tools are available to desktop retouchers. Just check out how it works on this 70-year-old photo.
Coming at you from Glyn Dewis is a video that incorporates a bunch of different tools in the service of restoring a damaged 70-year-old family photo of his. While ordinarily, this would be a quick retouch job to remove scratches and tears, in this particular case, Dewis wanted to be able to not only do that, but increase the resolution and make it an aspect ratio that would work for an 8x10" print as well.
The first tool Dewis uses should be pretty familiar to most photographers, and that's one of Adobe Photoshop's neural filters, the photo restoration filter, which does a decent job of most basic cleanup functions. As he points out, there are some other functions there to colorize photos, but for this video, he sticks to black and white.
The mind-blowing part came in how Dewis changed the aspect ratio of the photo. Using a tool called DALL-E 2, which describes itself as "a new AI system that can create realistic images and art from a description in natural language." Dewis was able to pop his photo into the system and have the AI fill in the missing background details quite convincingly, such that he now had enough headroom to crop into the aspect ratio he wanted.
I got so wrapped up in how this tool works. It's not perfect by any stretch of the imagination, but Dewis' use case is an interesting one for photographers.
The last step is a piece of software that's not as esoteric as DALL-E, and that's Topaz Gigapixel AI, an image upscaler that can increase the size of an image by 600%, since DALL-E is a bit limited in its output resolution.
With those three tools, Dewis was able to retouch his old photo with much less effort than it would have taken in the past.
What do you think of the results?
I misread the title and thought AI had actually colorized and restored the photo and was a little disappointed in the final result. That being said, I have no doubt Fstoppers will be featuring a video within the next 12 months showing exactly what I was expecting...AI colorizing. Actually, surely this has already happened right?
Actually, there are huge caveats to this "working" and not working". The image has to have just a little damage, in fact only a bit more than dust and scratches can handle. Any more and there is no chance. It will remove more important features than damaged ones. It won't work on any photo with missing pieces at all. As a photo restoration artist myself for 17 years this doesn't work. AI face enhancement only works if you have restored the faces properly in the first place and even then "myheritage" does it more naturally. The colouring is still terrible in photoshop and ill be using manual techniques to ensure a much better result and turning to AI for a good starting point. But not photoshop AI. I for one who works in this industry won't be using this as part of my daily workflow. It simply isn't good enough. As Patrick Hall said, within the next 12 months who knows.
It's amazing to see how technology can help restore old family photos with such incredible results. Glyn Dewis' video demonstrates how different tools, including Adobe Photoshop's neural filter and Topaz Gigapixel AI, can work together to not only remove scratches and tears but also increase the resolution and change the aspect ratio of a damaged 70-year-old photo. The use of DALL-E 2 to fill in missing background details is particularly fascinating, as it's an AI system that creates realistic images based on natural language descriptions. While the output isn't perfect, it's impressive to see how much easier and efficient photo restoration has become with the help of these tools. Overall, the results are truly stunning and offer a glimpse into the possibilities of AI-powered image restoration.