Think of a scene you’ve always wanted to photograph. You could go there one day – or you could synthesize it at your desk right now just by typing what you were thinking of into an artificial intelligence (AI) image synthesizer engine. Now take a shot of the moon with your smartphone? Impressed? You soon will be when an AI algorithm sharpens the detail to make it look like you used a ‘proper’ camera. AI is changing photography at a rapid rate and what the profession will look like in five years is anyone’s guess – and yet AI has been an important part of cameras for decades.

Welcome to AI photography – also known as computational photography and now ‘synthography’ – which comes with hard-to-resists advantages, unexpected pitfalls and surprising ethical considerations.

Here’s everything you need to know…

‍

1 What is AI photography? 

‍

AI is everywhere in photography. Recent developments are making it a highly controversial topic, with synthesized image generation now easily accessible. However, at its core AI in photography is nothing new. Also known as computational photography, AI is a bunch of algorithms that are trained on a dataset and identify patterns to help make decisions in real time.

It can be as simple and innocent as your camera identifying a face or an animal to adjusting the exposure to suit a particular scene, but as complex and controversial as a cutting-edge text-to-image engine that synthesizes images found online to create something entirely new and abstract. 

‍

2 AI photography: text-to-image synthesis

‍

Have you come across ‘synthography’ yet? In the last year or two, you’ve probably heard about or used AI image generators that turn text into computer-generated images. Services such as Midjourney, DALL-E 2, Stable Diffusion and NightCafe let you input text describing the image you’re after. They then use AI to recognize patterns and make predictions to generate photo-realistic images (and gifs, videos and 3D images) in various styles.

The results are often pretty crazy-looking to the extent that some think text-to-image synthesis is merely a passing novelty. You’re most likely to see them used in logos and in branding, but also on social media where ‘synthographers’ are posting some intriguing results – and without ever picking up a camera.  

‍

3 AI photography: fake selfies and photoshoots

‍

Another controversial use of AI in photography is the trend to use the new tech to create realistic and professional-looking portrait-style headshots from basic selfies, essentially for LinkedIn and TikTok profiles. Examples include FaceTune, HeadshotPro and TriyitonAI. The technology uses face recognition and AI image generation to create fake headshots and even expensive-looking photoshoot-style images.

It can also be used to see if an item of clothing looks good on you. Will it put headshot photographers out of work? Probably, but it gets weirder because there are already Instagram accounts that post AI-generated portrait photos of people who simply don’t exist. Even more confusingly, some of them are as impressive as they are intriguing. Is a new art form being born? 

‍

4 AI photography: deepfakes

‍

Perhaps the most insidious form of AI in photography is deepfakes, which are created to intentionally mislead. From inserting celebrity faces into X-rated movies to creating videos of events that never happened, deepfakes can take for the form of videos or photos. The whole concept revolves around inserting well-known faces into videos and photos to make realistic-looking fake media, though there are often giveaways.

Deepfake videos tend to be very poor quality, which isn't normally what you would expect from media that includes celebrities and politicians. They also often feature odd-looking eye movements and lip-sync issues. Deepfake photos tend to come via AI image generators, which are famously bad at reproducing human fingers (always count them!), teeth (which are often crooked) and, bizarrely, the number of arms and legs. Look beyond the famous face and you can usually spot a deepfake, though if you are in doubt then a surprisingly quick and effective way is to take a quiz compiled by Microsoft. 

‍

5 AI photography: shooting for the moon

‍

Smartphones often post-process images seconds after they’re taken. That’s what HDR mode is, among others. However, what if you take a blurry photo of the moon only for your smartphone to instantly make it seem super-detailed?

That’s what happens with Samsung smartphones, which use AI-based scene optimization technology to recognize the moon as the subject when in super-zoom mode. It appears to compare images to data it already has on the lunar surface (which, after all, doesn’t change much because the moon is tidally locked to Earth so always shows us the same side). There’s no image overlaying. An insanely clever use of AI or bordering on fabrication? It’s largely a one-off use of the technology and, besides, ‘real’ photos of the moon taken using DSLR cameras remain a lot better.

6 AI photography: post-processing and image handling

‍

Photoshop and other post-processing software have been around for years and have always been at the bleeding edge of AI in photography. Software like this uses AI to enhance image quality by reducing noise, removing unwanted elements and improving sharpness and contrast.

Within image organization software AI helps photographers quickly sift through large volumes of images to select the best ones, identifying specific people, animals and locations. AI-powered software for photographers is everywhere – and we couldn’t live without it.

‍

7 AI photography: in-camera

‍

AI is used in photography to save time, particularly in smartphone cameras. Face recognition allows your camera to automatically focus on your subject’s face or track a moving subject, adjusting other settings accordingly. Portrait mode blurs backgrounds for an instant ‘bokeh’ look while Night Mode kicks in when the sun goes down. We take all that for granted and it’s non-controversial, but it’s a classic example of how algorithms are used all the time in photography. 

‍

8 AI photography: ethical considerations

‍

Are images synthesized by AI engines legitimate? An unsolved issue in the current trend for AI image generation is the original source of all of these images, elements of images and backgrounds. The AI behind these computer-generated images are trained on datasets on millions of images, but which ones? Whose images are they remixing, collating and learning from? It’s not clear and no permissions have been sought.

Should the original photographers be rewarded, or at least acknowledged? It’s impossible to tell which images are being used by AI image synthesizers, so although it’s a huge grey area there’s little prospect of an easy solution that rewards hard-working photographers. However, it does beg the question of how the AI image generators of the future will work. If photographers take fewer new image and rely solely on generating them using AI then the datasets used by the AI will eventually age, look dated and lose their currency. On the other hand, it’s worth recognising that perfecting a text-to-image AI synthesis engine is complex, time-consuming and hugely impressive work that perhaps shouldn’t be judged negatively merely because it works so very well and so very quickly. 

9 AI photography: the future

‍

AI is advancing to the point that it will soon be impossible to tell the difference between ‘real’ photography and AI-generated photography. Pretty soon you won’t be able to trust that any face you see online is a real face. Articles on websites will use AI-generated images instead of paying for ‘real’ photos.

Is any of that really new? Headshots have been generously touched-up in Photoshop for decades and publishers have for a long time used online libraries stuffed with ‘creative’ images produced using software. The main difference is that while post-processing and creative image-generation was painstakingly slow in the past, it’s now instant and, consequently, much more popular and accessible. Will AI kill off ‘real’ photography? No, it will just change it – like it always does. The only real challenge is the speed at which it is doing so.