
I've spent the last week putting ChatGPT through its paces with photo editing tasks, and I'm delighted to report that it's spectacularly rubbish at it. Not just mildly disappointing or "needs improvement". Properly, wonderfully terrible.
And as someone who's watched AI encroach on creative territories with the relentless determination of Japanese knotweed, this incompetence feels like a blessed reprieve.
In fact, I'm starting to suspect that ChatGPT can't edit photos at all.
Why is it so bad?
So far, this experience has backed up unofficial information I've gathered about how ChatGPT works from trustworthy sources. That it doesn't truly "see" an image in the way humans do; it receives a text description and basically starts from scratch.
Which is why asking it to "make the woman's dress blue instead of red" often results in an entirely different woman in an entirely different dress, standing in what might charitably be called a location.
This isn't just a minor hiccup in AI's otherwise triumphant march toward creative domination. It's a fundamental limitation that exposes the difference between pattern recognition and genuine understanding.
I stumbled upon this inconvenient truth when I tried editing a photo for an online tech review. Here's the original, which is clearly quite dull and lifeless.
The best camera deals, reviews, product advice, and unmissable photography news, direct to your inbox!
I asked ChatGPT to brighten the image and make the colors punchier, and here's the new version it gave me.
As you can see, it hasn't edited the image but recreated it, albeit fairly accurately. It's created a brand new hand, which is holding the tablet at a different angle. The website on the screen features a different (presumably auto-generated) photo of the great Ed Sheeran. Even website text has changed; much of it is now unreadable drivel.
Here's a random selfie I fed into ChatGPT with a similar instruction to punch up the colours.
What it produced was a vividly bright image of someone who looks like me, but to my trained eye is very much not me (the mouth, especially).
It feels very much like someone was asked to digitally craft an e-fit picture of me based on a very good eye witness description. Which I gather, in very non-technical terms, is basically how ChatGPT operates.
Human capability
The short version is that ChatGPT claims to edit photos, but it actually doesn't. Meanwhile, the tools that professionals actually use to edit photos continue to evolve and improve.
Adobe's latest updates to Photoshop include AI-powered features that actually work because they're designed to assist human creativity rather than replace it. The difference is crucial: pro software enhances human capability, whilst ChatGPT's approach seems to be "let's start over and hope for the best."
So while the tech evangelists continue their breathless proclamations about AI's creative potential, those of us in the photo editing trenches can afford to be smugly confident.
ChatGPT might be able to write decent prose and generate impressive images from nothing. But ask it to make a simple edit to an existing photograph, and it reveals itself to be about as useful as a chocolate teapot.
Tom May is a freelance writer and editor specializing in art, photography, design and travel. He has been editor of Professional Photography magazine, associate editor at Creative Bloq, and deputy editor at net magazine. He has also worked for a wide range of mainstream titles including The Sun, Radio Times, NME, T3, Heat, Company and Bella.
You must confirm your public display name before commenting
Please logout and then login again, you will then be prompted to enter your display name.