.
(For A.I. poll, please go here.)
For years it has been in development, but in recent months, there has been remarkable breakthroughs in artificial intelligence — especially in the field of generative AI, that enables machines to create text and images.
The operative word is create. AI creates. It doesn’t provide facts, will lie with brazeness of a politician, and doubledown on mistakes (e.g. 1+1 = 3). But its artistic abilities are great. You can ask AI to write a James Bond script, paint Spongebob in style of Van Gogh, or compose an ode to Einstein in iambic pentameter.
So it can paint. Can it take photos in a style of a particular photographer. This is a tricky terrain since AI-generated images can still look surreal, dreamlike, and artificially lit — not a bad thing for paintings but can be tricky for photos, where you can end up with uncanny valley.
I set out to explore.
I use Henri Cartier-Bresson — partly because he was arguably the most famous and accessible photojournalist of last century (meaning there’s a lot of photos by him as reference material) and he has a distinctive style and visual elements that AI can learn to recognize. (Midjourney maintains a training set of photographers).
[February 2022] China. AI cannot generate legible text. I think lower righthand corner is supposed to be an eunuch, common in many of HCB’s China photos, but the figure is contorted. Faces are also contorted. 
[March 2022] Tokyo Market. Composition here is reminiscent of some of his photos but figures are definitely surreal 
[October 2022] Train Station, Europe. This one is saved by the fact that faces are not shown. The rundown train station and the crowd is more reminiscent of Yevgeny Khadei. Feet, perspective, and some of the details are wrong. 
[March 2023] Woman at Computer. This looks like a photo spread you probably could have seen in LIFE magazine back in the day. Significant improvement from a year ago. 
[March 2023] Woman at Computer. Likewise here. See details on the ground and the dust in the foreground, although background renders are a bit wonky. 
[March 2023] Africa, 1960. These last photos are probably the best renderings of faces. Once again, more James Natchway than Bresson, but impressive. 
[March 2023] Japanese sea food market, redux. Significant improvement against September results. Faces still not optimum. 
[March 2023] Paris Street. Great photos. Reminded me of Robert Doiseau.
.
So what’s the verdict?
Even the current photos don’t look that realistic, but it has made huge improvements in last 4-6 months. I cannot imagine what it can do in 2-3 months’ time.
Copying the style of one photographer seems to be the more challenging part. Unlike van Gogh or Leonardo or Turner, the style of a photographer can be hard to pin down, when many photographers may use similar framings, similar cameras and settings, and the same photographer will use different cameras and settings.
For me, I have tried to generate some of the photos above using these settings: “Leica M10 Monochrom. ISO 400, f/4, 1/125s, 35mm”.
.
If you like what I do and what I write, or simply wants me to write more, you can support me via Patreon. I had tremendous fun researching and writing Iconic Photos, and the Patreon is a way for this blog to be self-sustaining.
Currently there is a poll there on:
What sort of iconic photos / images from past eras you want to see? In which photographer’s style? Please go and comment. (https://www.patreon.com/posts/poll-iconic-by-i-80820823)
Proceeds mainly go to buying photography reference books and support me on my research (re: paywalled articles, trips to various archives). In addition to monthly addenda posts on Patreon, readers who subscribe on Patreon might have access to a few blog posts early; chance to request topics or to participate in some polls
Even if Patreon isn’t your thing, you can support by re-sharing, or tweeting about the blog or the specific posts on here. Thanks for your continued support! Here is the link:
Iconic Photos 