There’s been a bit of chat recently about how artificial intelligence will mean the end for photography because the AI generators are getting better and better. I got early access to the Dalle-2 beta which creates images (including photo-realistic images) from text. Type in “black and white photograph of a man jumping over a puddle in Paris in the 1930s” and it gives you the images shown above.
You get four high resolution images per text input. The five coloured squares bottom right indicate it’s a Dalle-2 image although you are not obliged to keep them on the image when sharing. Henri Cartier-Bresson this is not. It’s not even photography although it might pass for it*. It is fun though.
Here are three more prompts I tried:
black and white photograph of a man jumping over a puddle in Paris in the 2030s
photograph of a man jumping over a puddle in Paris in the 2030s
photograph of a man jumping over a puddle in Paris in the 2130s (because why not?)
* I said this isn’t photography. But why isn’t it? Some of them look like photographs. Where do we draw the line?
My preferred definition of photography is that it’s an image made in the real world with light and time, caught in two dimensions. This includes all lens-based photography but also cyanotypes and other camera-less images, Google Earth images and X-rays (because you can still have a photograph even if the light involved isn’t visible to the human eye). Other people have other definitions; what’s yours?
I think this technology might spell the end for cheaper stock photography. Why pay hundreds of dollars for commercial use of an image when you can get one included in your AI subscription? Especially when you can be exceptionally specific about what you want when you need something that you are unlikely to find in a stock photo library:
“large sculpture in a desolate field of a bird made out of green moss”
But it’s not the end of photography. The reverse, in fact. I think it is making us think about what makes us human, why we need to create and why it’s special to go out with a camera. We are more likely to find value in a “real life” photograph once we know what the alternatives are. We might even be moved to stop churning out the same-old same-old that we see on social media (because it’s easy for AI to replicate that). What can we do that’s personal, different, interesting, individual? What can AI NOT do?
Dall-E is pretty addictive. Here are some more things I (it? we?) created, not just photographs. It can be fun to try and reverse engineer what the text prompts were and you can find entire communities on Twitter and Instagram dedicated to this. The last one of the fire engines I created on the day it was 40 degrees here in the UK and we had fires by the side of the motorway. I just wanted to see how good AI would be at creating fake news. Pretty good I’d say: don’t trust everything you see online.