Photography and A.I. Computer generated A.I. Image of a bearded man by DALL-E

Photography and A.I.

The last year or so has seen an explosion in A.I. technology being reported in the media. What was once seen as science fiction, has, over the last couple of decades, become more prevalent and more widespread. Most people by now would have heard about Chat GPT on the news. This technology is exciting, but also raises serious questions how best to integrate this into our lives, or even if we should. While the pitfalls and positives are now being debated, I have been thinking about how this can affect our photographic practice. How will the photography world embrace A.I.? Will it be for good or ill?

Most photographers using Adobe products in the last few years may have come across their A.I. processing tools, which, in the right hands open many new opportunities for processing digital images. Not liking a bland blue sky? A.I. will now replace this with a beautiful sunset. Adobe have been advertising Photoshop as the ‘worlds most advanced A.I. app’ since late 2020, and for those invested in the Adobe ecosystem, it is hard not to be impressed even if those features are not necessarily part of your workflow.

I first encountered the power of what A.I. could do from being sent a link, showing at first glance, images typical of the style of Victorian photographs. As someone who has researched Victorian photographs extensively, they looked nothing out of the ordinary. However, they were all generated using the A.I. software Midjourney by artist Mario Cavalli. I was surprised by the accuracy of the images. It was only on close inspection of a few images did the inaccuracies start to show. This excited me, not only by what could be achieved using A.I., but also from the theoretical possibilities that A.I. generated images within a photographic discourse could open.

My first thought was that this is not a photograph. It looks to all intents a digitised copy of a wet-collodion photograph, but being an A.I. generated one, it is not even that. The surrealist painter René Magritte in his painting ‘The Treachery of Images (This is Not a Pipe), (1929) painted a picture of a pipe captioned underneath ‘Ceci n’est pas une pipe’. He was right, it was a representation of a pipe, not a real, tangible pipe.

A.I. generated images are representations of photographs, but without an original. A digital copy of a photograph that never existed: A simulacrum. While the concept of simulacra has been discussed since Plato, it was the French semiotician and theorist Jean Baudrillard in his book Simulacra and Simulation who theorised a simulacrum is not a copy of something that has existed but exists as its own truth. “Whereas representation attempts to absorb simulation by interpreting it as a false representation, simulation envelops the whole edifice of representation itself as a simulacrum” (Baudrillard, 1994). I believe this theory on simulacrum helps to define what A.I. images are within a theoretical context. Of course, there is much more detail that can be written A.I. and simulacrum, but that it for another post.

One line in Simulacra and Simulation stood out regarding A.I. images; “When the real is no longer what it was, nostalgia assumes its full meaning” (Baudrillard, 1994). Unpicking this, my own interpretation is that A.I. generated images, made to look like photographs that have no original, could become so widespread that we fall back on our own interpretations of what we believe the image is showing us. In other words, we will use our own visual experience and knowledge to fill in the blanks that are missing from the A.I. image and construct our own meanings or explanations around it.

The way most A.I. image generators work is from a text prompt. This is a set of instructions that the A.I. ‘reads’ then builds up an image generated from the masses of images already available on the internet. After playing around with prompts on DALL-E 2 for a while, the limitations are obvious but it’s still striking by the results the A.I. can generate. The skill is in creating the prompt. The more concise and detailed, the better the resulting image. Future skill with A.I. depends on people’s ability to construct the right prompts to generate the desired outcome.

This gets me thinking. From a photographic point of view, if images can be generated from a few lines describing what to show, where does that leave the photographer? Happily, I believe the skills, knowledge and experience of the photographer far outweigh anything that the A.I. can generate. Looking at the general photographic discourse, I can’t see A.I. ever being more than a helpful tool to help process digital images, rather than replace photographers. Unless clients start asking for A.I. generate images, but that is another matter.

However, if you’re processing a photograph to the extent of replacing large elements such as the sky, at what point does this stop being a photograph and start becoming ‘(digital) art’? This is a fine line that relies on the photographer to be open. I personally have nothing against quite severe and extensive processing: removing elements of the image, replacing others (such as the sky) or adding in other elements in order to suit a particular aesthetic or narrative. This is nothing new in photography, and something that has existed before the advent of Photoshop through photomontage. Oscar Rejlander is probably the most well-known Victorian photographer who practised photomontage. The likes of Peter Kennard’s political statements are testament to its continued power. While the likes of Photoshop make photomontage easier, the underlying principle is the same. Manipulating elements from different photographic (or other media) sources to construct a new image, often with a with a new meaning.

However, it should be made clear that this level of manipulation has taken place if it’s not immediately obvious. More so if parts of it have been altered by A.I. It’s not diminishing the skills of the photographer. Far from it, but adding extra knowledge for the viewer as to how the image came to be.

Not a photograph – Image generated by DALL-E 2 A.I.

The article on Cavalli’s images did mention the role that the photograph plays as a document. Up until the last few years, every photographic image that existed was, on a basic level, an image of someone or something that existed. If you saw a Victorian portrait photograph, you knew that at one time that person was alive.

A.I. generated images such as those by Cavalli disrupt this. It is not a painting, or even a photo-realistic painting, but it is using the tropes and styles of Victorian photography to construct images that have these particular aesthetic qualities: The sharp subject, deep out of focus background, specific B&W toning, vignetting in the corners. Our visual culture tells us what a Victorian photograph should look like. Without the knowledge that these are in effect ‘digital art’, these images could be interpreted as actual Victorian images of people who existed, which is where I believe the problem with A.I. generated images can lie.

Photographic history is littered with images that have been altered or manipulated to fit a particular discourse. From Abraham Lincoln’s head on John Calhoun’s body, to Stalin’s erasing of people who fell out of favour. However, these photos are well known, and the people existed – even if Stalin’s regime thought otherwise. The images may have been altered, but there is a narrative, a discourse that existed before the image was taken, and after it with the photograph entering the public sphere as a document. With A.I. generated images, there is no before narrative. Unless we know it is an A.I. generated image, what has gone before does not exist except in what the image shows us – or what the A.I. has been told to show us. As a viewer, we then create our own, incorrect, interpretation based on this information.

To illustrate this article, I used the free A.I. image software DALLE-2 to generate two images in the style of Victorian photographs. The results are quite believable but there are issues once the images are scrutinised closer. As A.I. technology improves and becomes more widespread, there could be situations where images are created to suit someone’s own narrative to which the generated image is now a ‘proof’, a false document that could then enter contemporary discourse. This has already started to trickle onto social media, with a few examples of people generating images with A.I. in order create false narratives. This has mostly been harmless with the author eventually coming clean to say the images were A.I. generated, but the potential use of this technology to document or evidence events that did not happen is terrifying.

Currently there are filters on A.I. image generators that will stop damaging images being created in the first place, but this wouldn’t stop innocently generated images being mistaken for people who’d once lived. It is up to the creator to do as much as they can to make people aware that the images are A.I., in a hope that the context around them comes from the fact they are generated by a computer. On a more positive note, it is conceivable that there could be competitions for A.I. generated images in much the same way they exist in the digital art world. Maybe Victorian-style generated images could be a category?

While a lot of the current talk around A.I. appears to be quite negative, I do feel that the underlying technology should be praised and marvelled. A.I. could help with medical breakthroughs and in that regard, it should be pursued. In our photographic world, A.I. is already here in the form of clever software, but as far as image generation goes there is still a lot to discover as well as control.

I personally feel this is exciting, but it is up to us as users and consumers to decide how this technology is used, and how far, morally, it should go. For photographers, artists and creators using A.I. software to generate or enhance our images, being open and honest about what has been achieved allows it to be contextualised on its own merits, within its own discourse.

Baudrillard, J., (1994) Simulacra and Simulation, University of Michigan Press, Michigan

Growcoot, M., (2020), Victorian-Era People Who Never Existed: These Portraits Were AI-Generated [online], Available from:

Magritte, R., (1929) The Treachery of Images (This is Not a Pipe), Los Angeles County Museum of Art