Since I'm not particularly keen on unemployment, I naturally follow these developments with keen interest. After putting Photoshop's "Generative Fill" through every conceivable torture test in 2023, it was time for Black Forest Labs' brand-new Flux engine. June 2025, latest generation, all stops pulled out.
For my test, I used the same photo of Dominika lying relaxed on a golf course green. To be fair, the comparison should be done under equal conditions — after all, I'm a photographer, not a scientist (though the lines sometimes blur).
The first thing I noticed: Dominika got "optimized." Even though that definitely wasn't part of my brief. The AI apparently had its own ideas about how a woman on a golf course should look. Sometimes her foot got twisted in anatomically questionable directions, her face manipulated beyond recognition, or at least the lighting changed so drastically that the result had about as much in common with my original as dog poop has with chocolate cake.
Optimized? Fine by me, if I ask for it. But when no optimization is requested, the AI doesn't just work counterproductively—it works dangerously.
The golf carts themselves? Actually quite decent, I have to admit. The AI understood what a golf cart is and roughly how it should look. But with the people on it, things got weird again. A male couple here, a mixed couple there, sometimes even a threesome on a vehicle definitely not designed for three people. My prompt said "couple," not "acrobatic troupe." But hey, you can't deny the AI's creativity.
Are the results photorealistic? At first glance, maybe. At second glance, definitely not. Could you fool people on social media with them? Absolutely. Is this the future of photography? Never.
After this test, I'm convinced we're still much further from reality with AI image generation than media coverage suggests. Of course, that's only if you look critically.
I'll keep watching these developments. Not out of fear, but out of professional curiosity.
