Artificial Intelligence is a powerful machine-learning technology that's starting to seep into our image editing workflow and it'll soon render some manual editing obsolete, but which will go first?
Though there's a lot of image editing software out there now that features Artificial Intelligence (AI), it's not yet been fully realized by software companies. Skylum has put both feet firmly in the ring by offering what they dub as "the first image editor fully powered by artificial intelligence" and others aren't too far behind. But there's still a long way to go when it comes to perfecting images using machine-learning tech alone. However, there are seven specific edits that we think could easily be replaced in the next few years, so read on to find out what they might be.
Cutting Out Subjects
This one's been around for a little while now, but it's set to get incredibly sophisticated. There are plenty of online sites and image editing software that now offer autonomous subject cut-out. It uses machine learning artificial intelligence (AI) to scan an image, decide where the main subject is and draw a path around it ready for masking. If you haven't used this before, take a couple of seconds to explore it in your favorite image editing software (if it has this feature) or try some of the tools online. For the most part, it does a good job.
However, things start to go awry when you're working with more complex shapes and textures such as hair, trees, or anything with naturally complicated edges. That's not too much bother since most retouchers are used to working with this kind of issue, so we make a selection and then refine it before moving on with our edits. But AI is still in its infancy in image editing.
Give it another five or ten years and we should see massive improvements where complex edges are no longer a factor at play in the cut-out. In fact, taking it to the extremes, we should be able to use a voice-activated assistant to cut out our images. "Alexa, please cut out the woman, the weeping willow tree, and the car passing in the background." It may sound far-fetched now, but this is already somewhat doable. Look at Adobe Photoshop and you can use the Object Selection tool to draw around particular subjects. I could do this myself by drawing three separate boxes around the subjects and having the software take care of the cut-out for me, or, if we set up the software correctly we could have the assistant act as our mouse cursor.
Colorizing and Restoring Old Photos
Another realm of AI-powered editing that's reared its head these past few years is colorizing old monochrome photos. Machine-learning can even refine facial features and remove folds or creases on antique images. Again though, it's not perfect. Colors often have to be helped along with manual input. Masks still need creating for things such as skin tone, clothing, and different textures. This is an area that has massive potential for growth though, as these kinds of edits are time-consuming when done manually.
This one's an old one too, using AI to remove distracting elements. Say you have a cityscape scene but you want to remove a couple of people and some signs from the background. We have the editing technology to do this manually, or via AI (just look at Google's new Pixel 6 and 6 Pro — that feature is built right into the phones). But it falters when there are intersecting elements, some of which we want to remove and other bits we want to keep.
For example, someone's hand in the foreground cutting across a car we wish to remove behind - we're likely to see parts of the hand disappear as software struggles to decipher which bits to remove. Soon though, machine-learning will be so intelligent it could remove distractions in the foreground or background, intelligently fill areas required, and extend its reach to reducing and removing light flares by drawing in sections that are overexposed and rebalancing the existing image data.
Keywording Images for Cataloguing
Open up Lightroom and type the word "cat" into the address bar and you'll see that our image editing software is already starting to scan our back catalog for subject types. For now, it's limited in scope to very obvious and specific topics, but soon it'll be able to correctly identify a huge array of things. Boring keywording will become a thing of the past as we rely on the software to automatically return results based on what we search for. "Mountains", "Italy", or "Lamps" will quickly return every photo you have in the library with those things in. We'll be able to include multiple search terms too, "cats on grass" or "the constellation Orion" will make it increasingly easy to find our favorite images.
Looking forward I'm gonna go out on a limb here and say that machine-learning technology will become so efficient that it'll be able to study previous edits we've made and edit future photographs to mimic what it thinks we'd do. I mean, that's what Artificial Intelligence is all about: informing the machine with data until it has a big enough input that it can start making assumptions itself based on what it's learned. So why would making image edits be any different? There are already types of this floating around, where styles are borrowed from great artists and applied to our images, so why not from ourselves? All it takes is a few hundred of our edits of different genres plumbed into the machine and it'll edit future shots based on how we would want it edited.
Manipulate Depth of Field
We already have the ability to produce shallow depth of field effects through the use of blurring and intelligent cut-out technology in our smartphones and image editing software. However, with AI technology that already exists, I can see this being expanded upon further and in the opposite direction. Foreground and background content in photos will be enhanced to be made clear. This may be combined with camera technology so that it's hard-wired into the shots (this already exists in some cameras such as the Lytro) meaning you can take a shot and choose any kind of depth of field or focus point.
While this technology is out there right now, it's not super sophisticated and it hasn't made it into the mainstream market. As soon as manufacturers introduce this technology to all cameras it'll automate the photo-taking process, meaning an image can be captured and the focus point and/or depth of field can be determined after the fact by art directors or another creative.
We have a limited ability to up or downsize our raster photographs currently through image editing software, but there will be a huge shift in this technology when AI latches hold. We can see this now in many AI-powered image editing software such as Luminar AI and a little bit in Photoshop's Neural Filters. However, taking this technology to its ultimate endpoint we should be able to resize (or should I say, upsize) an image so massively that it won't matter what resolution we shot at.
This has the potential to cause a paradigm shift in how manufacturers currently sell their cameras. One of the first things we're told is how many megapixels the camera has, because this gives us an idea of how much detail it can capture. But if we can automatically resize images, as long as we have a certain amount of data we won't need the biggest number. This in turn can affect how cameras are made as, generally speaking, if two image sensors of identical size are compared then lower resolution image sensors produce less image noise than their higher-resolution counterparts. If we only need XX megapixels to then upsize using AI, then cameras can come loaded with that specific amount and we can really start breaking some low light shooting barriers.