Thursday, October 6, 2022

DALL-E - the photographer

DALL-E and other AI art programs have been causing both excitement and consternation in the art community - everything from “Wow! that’s amazing” to “It’s the end of art”. A cold, wet Wednesday felt like a good time to experiment with DALL-E and see what it could do for photography.

Back in 2010, I got up before the lark (which isn’t that difficult when there don’t seem to be any larks where you live) and went to New Brighton to take some sunrise photos. Among them, I captured this shot of the New Brighton Pier.

So, I thought this might make a good DALL-E test, and asked DALL-E to give me “A photo of New Brighton pier in New Zealand at sunrise taken with a DSLR camera”. (My first prompt did not include “in New Zealand” which was when I realised that there must be other New Brightons with piers around the World, as the pictures looked nothing like our pier.) Anyway, once I included "in New Zealand" this is what DALL-E provided:

At the small size DALL-E produces, it seemed like a pretty good attempt. The perspective was rather different to mine (to be expected) but the colouring was totally believable and the pier was the correct style and orientation to the sunrise.

Perhaps I could try changing the camera and, as I had been doing quite a bit of pinhole photography lately, tried this prompt: “A photo of New Brighton pier in New Zealand at sunrise taken with a pinhole camera using Kodak Portra 400”. This was the result:

Again a different perspective, but DALL-E managed to replicate both the dreaminess of a pinhole camera and the pastel-like colours to be expected from Portra 400. What if I changed the film to Ilford HP5?

Again, exactly what might be expected from Ilford’s iconic black and white film. I noticed that DALL-E didn’t simply rework an existing picture with a new camera or film; it started afresh producing another four images of the pier each time. I simply chose the one of the four that I liked the most.

So, what are my conclusions? I think that both camps - the wow! camp and the end of the world camp - are right, and they are both wrong. 

There is very much a wow factor when you see the images. That a computer which is half a world away from New Brighton beach can produce such a believable result is actually quite amazing. In certain limited settings, one could probably get away with presenting a DALL-E image as the real McCoy. Certainly, the DSLR picture is quite believable at first glance.

BUT the devil, as they say, is in the details or, in DALL-E’s case, the lack of details. DALL-E has produced a good impression of New Brighton pier, but the details are all messed up. There are bits there that shouldn’t be and bits missing that should be there. It’s DALL-E’s IMPRESSION of the pier gleaned from many photographs, but it’s not an actual image of the pier. It’s an impressive effort but falls short - a bit like a kindergarten drawing drawn from a child's memory.

DALL-E is good with style though. It understands the style of a DSLR image, a pinhole image, Portra 400 and Ilford HP5. From other DALL-E images, it is also clear that it understands the style of Van Gogh or Botticelli, of cartoons and illustrations. So DALL-E is good at putting a veneer of style onto an image. But not so good at producing anything more than an impression of a thing or place. 

One thing’s for sure, as lazy as I am and as impressive as DALL-E is, it is not ready to replace getting up at stupid o’clock in the night to go and photograph the sunrise at New Brighton beach. Photographers can sleep easy, DALL-E won't be replacing their camera any time soon.

No comments:

Post a Comment