“The future of photography is defined by AI image generators”

Photographic artist, digital consultant and university lecturer Boris Eldagsen on his experiments with AI imaging processes, the future of the photographic guild, the question of why everyone should actually get to grips with artificial intelligence and why he welcomes this disruptive technology with open arms.

Mr Eldagsen, since last summer you have been experimenting intensively with imaging AI processes. A good six months later, you are considered an expert, give lectures, organise workshops. Is this rapid “becoming an expert” emblematic of the rapid development of the technology as a whole?

You could say that. Artificial intelligence is sprinting ahead at seven-mile pace, with new, better-performing, creativity-inspiring features appearing in the relevant programmes practically every week. The pace is breathtaking.

Let’s take a quick sketch of the development steps so far.

The technology started with GANs, simply put, two computing units that approach a desired result in a trial & error process. That was impressive, as sites like “This person does not exist” showed a few years ago. In the meantime, however, we have progressed technologically. Thanks to the so-called diffusion models, we are at “text to image”: language becomes image at the push of a button, so to speak.

In simple words: What is the technical process?

I enter terms – so-called “prompts” – which can range from simple words to descriptions of feelings, and the AI first resolves a suitable image from the training arsenal of the AI programme into white noise and then builds a new image along the prompts.

Much of your AI work seems nightmarish. One is reminded of surrealist films such as “An Andalusian Dog” or David Lynch’s “Eraserhead”, sometimes one sees indefinable organic structures that look like Hieronymus Bosch got his hands on Photoshop. You want to look away, but you can’t…

Fascinating, isn’t it? I still come from the time of the darkroom and have worked a lot with alienations of analogue image material. But I probably couldn’t have generated these images with any of the existing technologies.

Formulated in a generally understandable way: How did these images come about?

I subjected various AI programmes – including the commercial programme DALL-E 2 or the open-source programme Stable Diffusion – to a creative stress test. I tried to test the limits of the programme’s internal censorship, for example by using rewrites of blocked words to get the AI to generate images that violated its own guidelines. At some point I was blocked, used a new account and worked with nonsense prompts – such as misspelled words, semantically nonsensical ones or nested sentences in which the subject and object could no longer be distinguished. In the meantime, I understand creative work with prompts as an art in itself.

The fact that a machine and not a human being generates image content doesn’t seem to frighten you – on the contrary: you welcome this disruptive technology with open arms. Why?

It’s a lot of fun. The results – if you invest time and are smart about it – are striking. New creative spaces open up.

Don’t you miss interacting with the natural world and people?

I still go out, of course. For example, I regularly take pictures at night on Museum Island, always in the same place, and it never gets boring. I also enjoy working with other people who can do things I can’t -that’s inspiring. It’s similar in the interaction with the AI. It’s a real collaboration, you could say a duet.

AI as an inspiring playground for photo artists immediately makes sense to us. But what does the technology mean for applied photographers? To what extent is their business basis endangered in view of “non-photographic” business models that can thus be occupied by lateral entrants?

To put it simply: photographers who are close to documentary photography – news or wedding photographers, for example, or corporate photographers who portray business executives and companies – can sit back and relax. Their services cannot be replaced by any machine in the foreseeable future. In my view, everyone else has to come to terms with the technology.

Assuming a 50-year-old photographer is willing to learn the technology, the question remains: will he be able to develop a business model from it that will secure his livelihood in the future?

One thing is certain: All those who understand something about image composition, about learned photographic perspectives and about image communication in general are a decisive step ahead of all the others. Those who know about the effects of focal lengths, aperture, depth of field, who have knowledge of art history, can assemble their own style. AI can be a wonderful inspiration tool. On the other hand, I had another experience: a few months ago, at a BFF event, I juxtaposed automotive shots taken by BFF photographers with AI-generated car images. The result: very few people present were able to tell which pictures were photographed and which were AI-generated.

How will AI change the job market for image creators?

One thing is clear: the market will be shaken up, some job profiles will disappear, many new ones will emerge. By the way, this does not only affect image creators, but cultural creators in general – poets, musicians, dancers, even teachers are affected. But no one can predict exactly what will happen at the moment, the development is simply too dynamic for that. I find it remarkable that my AI workshops are increasingly being requested by publishers. My guess is that it’s mainly about potential cost savings. In the past, if they said we wanted this cover picture, but in green, then as a photographer I might have received a fee of 1,500 euros. Today, the AI produces it for me in no time at all for three cents. No wonder stock agencies are showing a strong interest in this technology. Model releases can be circumvented, even the “localisation”, i.e. the ethnic and linguistic adaptation to different cultural areas or the fact that you find many different motifs but often the same faces in stock agencies, can be solved in this way. This year will bring many new things that we can hardly guess at yet. Many new start-ups are emerging and a lot of money is flowing into this technology. The next step, which Google, Meta & Co are already working on, is text to video and text to 3D.

A prospect that is as exciting as it is scary. Where do you see the opportunities for visual creatives?

AI opens up completely new creative spaces. Fashion or car photographers can freely choose backgrounds, I can merge portraits of two people and thus generate a new “human being” who no longer requires a model release or a fee. I can use “inpainting” to replace one face with another or “outpainting” to have existing image content “extended” credibly at the edges by AI. The connection between chatbots – text-based dialogue systems for communicating with machines – and AI image generation also offers great potential. An example: I had a ChatGPT write a text about the hitherto non-existent art movement “Neuroplastic Expressionism”, shortened the generated text down to a few striking prompts, entered this into DALL-E 2 and thus created images of which I had no idea how else to produce them. The AI surprises me again and again – and not only me, by the way: I did an experiment, for example, and submitted an AI image in wet-plate aesthetics and ended up on a shortlist three times. In doing so, I did not violate any of the competition statutes. The jurors are simply not prepared for such submissions.

How do you assess the danger of deep fakes and opinion manipulation?

This is a problem and it will become bigger, also with regard to public opinion and democracy. Credible “alternative” images of war crimes can already be produced. The truthfulness or evidential value of images is thus finally tending towards zero, and in my experience as a former digital marketing consultant, the forgers are always one step ahead of the forensic experts. On the other hand, I am hopeful that AI can help to can help to find solutions more quickly in the areas of climate protection, drug development or social structures. And in the field of pornography, AI could produce new content without having to exploit people.

Will AI also change the language of images?

I assume so. At the moment, a lot is going in the direction of fantasy, most of it seems slick and quite mainstream. But I think that new, interesting visual languages will also emerge. It is the task of art to act as an innovator here.

And what advice would you give to newcomers? Which programmes should you look at first?

Personally, I started with DALL E2, but now I also use Midjourney and the open-source software Stable Diffusion. The former is fairly easy to use, but limited in what it can do. The latter almost requires coding skills, but is limitless in its functions. Moreover, I am sure that a multitude of new solutions will emerge very soon. Everyone has to try out for themselves which model or combination of models fits best.

INTERVIEW: Peter Schuffelen