Artificial intelligence (AI) continues to revolutionise various industries and is making significant contributions to the field of imaging. In the realm of photography, AI is being used to automate photo editing, enhance image quality, and assist with creative tasks, such as adjusting composition. This technology has empowered professional photographers to streamline their workflow and achieve new levels of creativity. The fusion of AI and photographic imaging holds great promise and the potential for creativity seems endless…or is perhaps the biggest threat we’ve seen to the future of photography, depending on your point of view.
Using AI to enhance an editing process
In photographic organisations, there has been much discussion, over the last months of 2023, about the legitimacy of using AI in photographic society competitions. AI up-sizing, noise reduction and sharpening software has been available for some time to enhance our existing photographs. The resulting images have been accepted, without question, as being legitimate for competition entry in photographic clubs and organisations. It is also generally accepted that the photographer still owns the intellectual property rights and copyright of the work.
Fusion of Photography with Generative AI
Now we have new generative AI processes appearing in our editing software, such as generative fill and generative expand in Photoshop. Perhaps in this category, we should include other AI based editing, like Photoshop’s Neural Filters, which can change expressions and eye direction for example (though not very well usually) and landscape filters that can greatly alter the look of the original photograph or blend more than one landscape together. Even in our phones, we can produce idealised and fantasy portraits, change the depth of field of the original and erase unwanted objects.
Generative AI processes can add entirely new areas to the images that we have already captured. This new content blends in with the existing photograph, but is not photographic. I’m going to refer to this as a fusion of photography with generative AI (GenAI). The question then arises, “how much of a given image is actually photography and how much is generative AI, so how legitimate is it to call the work all your own for competition purposes?”.
- Does 99% photography and 1% AI make an image legitimate for use in photographic competitions?
- Does 1% photography and 99% AI make an image legitimate for use in photographic competitions?
- Somewhere in between? Where in between?
- Who owns the copyright and IP of this work? Can the work be copyrighted at all?
- Can it’s use be controlled and if so, how?
At my photographic society, we have outlawed the use of GenAI in our competitions for now. Whether a section devoted to this might be envisaged in the future is, well, for the future. Beyond the confines of our club, there are many other competition organisers in the amateur and professional photographic world, who will need to decide how much GenAI is to be allowed (if at all) and how this would be regulated, for example by random checking and asking for the original photographic files of winning images. Many people agree that trust is not enough.
Generative AI – creating realistic artwork without photography
Let’s leave those awkward unanswered questions for now. Let’s just see the process simply as one that creates art and enjoy it for what it is, because creating artworks is not only about taking photographs or entering competitions. For anyone who likes to create, there are many other visual arts processes with which we can engage. Generative AI is just one of them and it can create whole images that look quite realistic but are not photography.
How it works
Generative AI programmes work on a variety of “models” which are trained on pre-existing artwork…millions or billions of images. A model uses complex algorithms, which allow it to recognise patterns, for example text inputs, called “prompts”, with the aim of creating entirely new images from scratch, or an original photograph can be uploaded as a starting point.
The computer user inputs “positive prompts”, which are “the things that you want in the image”…and “negative prompts”, which are “the things you don’t want”. Once it has its text prompts, the model can then produce entirely new artwork, photorealism if you want that, comic-book, cartoon or anime, or something that looks like an Old Masters painting. If you want to create photorealistic images of people, you could start with a model that is geared towards creating photorealism. There are lots of models around. They can be used to create entirely new data or change and augment existing data. The quality and content of the text prompts greatly influences the output…even changing one word or moving a comma can make a big difference to the finished image.
Some text inputs are very effective, others seem to be ignored. I am still learning the skill of text input. It is an iterative process where the first image created may be not close to what the “producer” of the image wants. Edit the text, add in a preset (you can make them yourself) and soon your prompting skill develops enough to achieve the image you want, something that you have helped to craft…of course there is an element of randomness in this as well, which can make it all the more exciting.
I stress again that it is important to understand that a totally generated AI image is not photography at all, even though it might look so much like a photograph that most people can’t tell. Forget photographic competitions and find another use for it!
My First Foray into AI Imaging
On 23rd December 2023, with the increasing accessibility of Open Source AI-powered imaging tools, I decided to embark on my first foray into AI imaging. Initially, I was curious about the possibilities it could offer. I decided that the only way I could evaluate it would be to plunge in and get my feet wet.
The first couple of days, I used Stable Diffusion XL (SDXL) on my android tablet with output generated “on the cloud” and produced lots of images like those in the gallery below.
With SDXL and a user interface now installed on my desktop computer, I am able to take the production of images much further very quickly and generate them locally, on my hard drive. SDXL is Open Source software, but this article isn’t so much about software as the results. There are lots of different packages out there, from free ones to those for which you pay a subscription and they will inevitably produce different looking results. Some popular ones are Midjourney, DALL-E3 (which is also used in Bing Image Creator) and Firefly in Adobe Photoshop.
The first images I produced were very stylised, with porcelain-smooth perfect skin. But, with more prompting and adding other moderating software, skin can look much more realistic, with pores and freckles or wrinkles. But how are other people going to respond to AI generated images and how much does it matter?
Appreciation of AI-generated Art
Some people believe that AI imaging it isn’t even art… but it begins with the human-created art of the photographs and images on which the software has been trained, then there is the skill of the software engineers, who have developed the models, and there is the imagination of the “producer” (I, me, myself) who has directed the software, via an iterative process of text instruction, until the desired picture is produced. Somewhere or another along the line, human imagination has been involved. Content, styling, composition, lighting and colours have been directed by the user and that is the basis of art, surely? Regarding ownership, the UK has not decided yet if AI images can be copyrighted but the US won’t copyright AI because it is machine generated.
Do we prefer human-created or AI-created artwork?
Regarding people’s attitude towards AI imaging, even among the people who say that the images are beautiful, there is sometimes a degree of negativity when they realise that the images have been produced without an initial photograph…especially people from the photographic community. It is perfectly natural that there is a degree of anti-AI feeling around. That will continue…as there is still some anti-photomontage feeling around, even after three decades of it appearing in our photo-competitions. In searching AI information on the web, I came across some studies that have been done which found the same bias against AI.
Two studies were carried out at an American University, to compare attitudes towards human-generated and AI-generated artwork. The methodology was rigorously examined and passed by the University ethics department and statistical testing found the results to be highly significant.
In the first study, about 150 subjects filled in a questionnaire to measure their attitudes towards 30 images, chosen at random by the researchers. Attitudes were measured on standard 5-point Likert scales. The images were first randomly assigned to two groups, 15 in each group. The images in one group were labelled as “human-created art” and the other 15 were labelled “AI-created art”. The subjects of the pictures were varied, with some realism (representational), some abstract, in each category.
In fact, and this is most important, all 30 images were totally AI generated, but the subjects of the study were never made aware of this.
The results showed that there was not a lot of difference between the aesthetic appreciation of both the human-art labelled images and AI-art labelled images. In fact, some AI-art labelled images were thought to be more beautiful than some human-art labelled images. However, images labelled as human-created art were marked higher than AI-created art images in several other categories, with the art labelled as AI-created being rated as less valuable and less profound and less meaningful than the images that the participants erroneously believed were human-made art.
A follow-up second study was made, again with about 150 participants, replicating the first study and again all 30 images were totally AI generated but half were labelled as human-created. The researchers added some extra questions around the subjects of emotional response, perceived narrative, the effort and time they thought had gone into creating the artwork, the value and quality of the artwork. Again, participants “on average” rated what was presented to them as human-created art more highly than what they thought was AI-created.
The results of both studies showed a significant anti-AI bias/pro-human bias. This preference was “particularly evident for criteria that communicated deeper meanings of the art”. It was as though the subjects inherently believed that human-created art is more deeply meaningful and more valuable and that it reflects actual human experience and creativity.
Because of the rigorous nature of the studies, and relatively large sample sizes, it is likely that the views shown will be replicated in the wider community and this prejudice and anti-AI bias will, I am sure, continue to affect people’s attitudes as AI-imaging becomes more and more prevalent. You can read the full research paper here.
Having posted a few of my GenAI images on Facebook, it became clear very soon that many people thought they were beautiful, but did look like AI, and some immediately warned “they can’t be used in photographic competitions”. Well, yes, I do know that…which is why I labelled them AI in the first place. But then I mischievously posted an image and didn’t say if it was photography or AI, so people had to ask. Here it is, “Old Woman with Cat”. Be honest, if you hadn’t been told it was entirely AI generated, would you really have suspected it? and if your answer is still yes, I would caution you that AI is only going to get better at fooling you. AI does have its limitations though…it doesn’t do eyes as well as it should and it struggles with hands and fingers…but I am sure we will see improvements in these areas over the next months.
My conclusions…as a photographer, I do not want to see AI generated works in photographic competitions, because it isn’t photography. I am unhappy about some “AI and photography fusion” too. A friend, who is a creative photographer wrote recently that he sees “2024 as the year of AI, but not in a good way”. I agree in part with this sentiment.
However, I do find the process of creation of “never to be called photography, totally AI images” fascinating and I enjoy the way that I can influence the output to the extent that I have a final image very close to my imagination…or even better! It is just another interest in addition to my real photography and the occasional bout of not very talented painting. I can see ways in which I can use the artwork that’s created, for example, as source material to influence my creative photography, to make illustrations for my history talks, to make prints and greetings cards and to share on social media.
This foray into the imaginary world of AI has been instructive, exciting, absorbing and dangerous and I will certainly be doing more. It’s dangerous, because I can see what a threat it is to photography and to professional artists. But, increasingly, I hear a minority of photographers telling us to embrace AI and whatever your stance on this, we know it’s here to stay.
Will AI kill off photography? No, but it will certainly have an effect. In the 19th century, the French painter Paul Delaroche declared, in response to the invention of photography: “From today, painting is dead”…but of course it wasn’t!