AI may split the camera market

Another day, another announcement of what artificial intelligence can do with images. At this point, AI can select the correct exposure, tone curve, white balance and colours, remove reflections, reduce noise, guess depth maps and add background blur (“bokeh”, depending on definition), upsample while adding detail, and aesthetically crop. It has come a long way towards creating credible 2D depictions of objects from learnt patterns and textures, and it can aid in procedural generation of 3D objects as well. Procedural generation, whether with or without AI, is now a part of many computer games as well as motion picture production.


These kinds of automatic manipulations may be welcome to the amateur photographer looking simply to present a good-looking rendition bearing resemblance to what he or she saw. But there will equally be a counter-trend of authentic photography, in which dirt, distracting objects and other small imperfections are used as proof that an image represents reality.

Press photographers will more than ever be asked to adhere to professional standards of representing a true scene, and some artistic photographers will also uphold this as the marker of true purpose in their work. In their view, any AI features will have no place in a professional camera, and there may therefore be a class of camera that completely omits such features. Food for thought: might some hot pixels actually be part of the scene? Is the original, non-auto white balance of a scene informational – for instance, should images taken under ambient tungsten light look a bit yellow?

My prediction would therefore be that we will mostly see AI features added to entry and mid-level cameras, while “pro” cameras will be optimised for other aspects, such as shutter delay, frame rate and buffer performance. They may emphasise noise suppression through conventional means, and high resolution to the extent that these two can be reasonably traded off against each other.

We will see some kinds of processing declared valid, such as super-resolution from continuous shooting on a relatively static subject, or dedicated super-resolution capture modes, while others will by the aforementioned group be frowned upon. Currently, many of the most advertised AI methods will fall into the second, frowned-upon, category.

On the other hand, more photo agencies than before will run image analyses based on adversarial learning (an AI technique) in an attempt to detect falsifications in the submitted images. This will be an effective deterrent since even if the photographer submits a well-forged image, possibly using a counter-trained AI, his or her files will be retained by the agency and may be re-analysed by a more advanced AI in a future timeframe. In spite of this, some individual photographers will continue to don dunce’s caps and quite possibly bury their careers after being called out for excessive manipulation.

And lifestyle and fashion photography may in the medium term benefit from cameras that output perfectly post-lit, blemish-free images that need merely be reviewed for artefacts, with a concurrent gradual transformation of the role of the image editor into one more like an image curator. But will such images remain relevant to the world at large? Will imperfections make a come-back even in lifestyle and fashion photography?

Time will tell, but the social media pitchforks certainly won’t go away. Only the sterotypical, narcissistic selfie will fade, rendered meaningless by an arsenal of AI tools that will allow anybody to post a perfect insta.

One comment

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.