Any cursory glance at my photo albums will show that I like to take photos of railway stations. I adore railways, I appreciate the architecture of the spaces they create, and I enjoy shooting them as a photographic subject. What I don’t enjoy is photographing people, especially when the stations and trains themselves are my intended focus.
It is therefore no surprise that Adobe Photoshop is what has finally lured me into trying any sort of ‘AI’ tool. Generative AI is now baked into its digital canvas, and so — after cautiously checking the terms of use — I decided to try this new feature out on a snap I took a few days ago, of a recently refurbished train departing Stadion t-bana.
I had to manually intervene from first click, as the ‘AI’ tool tried to be smart and auto-selected part of the carriage door — presumably down to some suggestive combination of focus and framing. But a quick lasso tool around one of the figures on the platform set it in the right direction. Time to try and remove the person with the backpack; see if I can get a less distracting composition:
And lo, the figure is gone! An impressive amount of station background is now made visible, and the composition feels more proper — however it took me a moment to realise the person with a yellow backpack had not been removed, but instead replaced with someone less obtrusive. Then it struck me: where did this new person come from?
Photoshop’s tool goes a step further, and provides some variations for each of these outcomes — presumably so we can tune for a more desirable effect. The middle thumbnail presented a different imaginary(?) figure, dressed in similar clothing. But the rightmost variation drew my eye because it appeared to add some non-existent signage to the station roof. What was going on there?
Well, as glimpses into the generative AI future go, this was about as unsettling as I had expected. This variation saw fit to install a restaurant bug zapper, which is a peculiar choice. The thumbnail originally struck me as interesting because I thought it might depict a member of cleaning staff, however on closer inspection it is at best an abstract suggestion of a typical cleaning cart. That, or a 5-dimensional pushchair.
It feels like quite the journey, to have gone from “hey, the tool works!” through “wait, it concocted people out of thin air” to “what the hell is that?”
I particularly ‘appreciate’ how the final image deftly tricks my brain into seeing something more legitimate, so long as I do not look directly at it. Are outcomes like this one being passed by AI models precisely because they blend in more? I imagine I could adopt this suggested figure into my photograph and that most other folk would subconsiously regard that as a cleaning cart, like I did. But oh that second glance is potent and cursed.
So, generative AI is uncanny and weird. I think I’ll be keeping it out of one of my more precious creative pastimes for a while yet to come…