AI, Art and editing pics

The last few days, I’ve had a lot of discussions about AI image generators and art. With tools like DALL-E, Midjourney, or Stable Diffusion, you can create images from just a short description and some parameters - though I think “AI” is a misleading term here. These algorithms aren’t intelligent or sentient; they’re more like remix generators, fed with existing knowledge (the training data). Since the beginning of the month, I’ve had a Midjourney subscription to test one of the generators. I’ve used it often to get a feel for the process and its potential. The surreal images and the open, social nature of Midjourney are really interesting and feel unique, but after a while, I’m not sure what to think about these image generators if I want to have a sophisticated opinion. One thing is for sure: they are a gamechanger.

What’s interesting is that some arguments in these discussions remind me of the dawn of photography and the old debate about whether photography is art.¹ Some of the same arguments are still happening in the photography scene today.² It’s a weird iteration of old topics, and I think there are many biases in the discussion. The longer I look into it, the more it feels like a mirror of human experience.

My biggest concern (and why I’m reluctant to post the images I’ve generated on my site) is the question of copyright. Stable Diffusion, the open source variant, made this clear when they published their training data. In a very interesting article,³ it became clear that almost none of the artists in the massive dataset gave consent.

Nearly half of the images, about 47%, were sourced from only 100 domains, with the largest number coming from Pinterest. Over a million images, or 8.5% of the total dataset, are scraped from Pinterest’s pinimg.com CDN.

I don’t think it’s any different with DALL-E or Midjourney. Some pictures may have an open license, but the crucial point is the datasets provided by other companies. We’ll see some very interesting court cases about this, and as long as there’s no good regulation, I’m reluctant to use an image generator outside of private use. Companies like Midjourney may give me a license to use the generated image, but no one will help me if an artist (or a shady law firm) sues me for using an image generated from their work in the training data.⁴

I also doubt the “uniqueness” of the image results. For example, with some prompts - like “snow queen” - the results were just Elsa from Disney’s Frozen or iterations of the same dark fantasy snow queen. The training data didn’t seem to have much tagged with those words. With that in mind, I have a hard time claiming the algorithm generated new art, especially when I start seeing identifiable things creeping through and can guess the basis data. This is a big problem if you start reselling the images for profit.

So what would be a good solution? I really don’t know. Given the nature of these things, attribution is probably impossible (though DARPA and Nvidia are working on “forensics tools” to detect this⁵). I’d prefer to see artists at least give permission, and if someone uses it commercially, there should be a cut/royalty/patronage - because in the end, their art and ideas were used in the training model. I could also imagine a protective system that artists can include in their work to prevent it from being used in systems like Midjourney.⁶

Aside from the legal issues, it would also be helpful to be able to research the training data to check for biases.⁷ I already have my own bias in assuming the model or dataset actually has properly labeled imagery and good training on the words I use - which is simply not the case. Hours and hours of parameter tweaking and word changes are gone just to get the result I wanted. On the bright side, I’ll be a walking thesaurus by the end of the month.

But is it really art when I generate a good piece with just a fitting prompt, or is it just exploitation of artists? I think that really depends on your viewpoint.⁸ There’s no good definition, and humans have debated “What is art and what is beauty?” for thousands of years. Every time a new technique arrives, the goalposts move. People much smarter than me have made careers out of examining that question. So here’s my humble opinion: Personally, the results from the generator are not fine art for me, even if they look like it. There are still visible artifacts - like the eyes, hands, or the concept of a person holding things. It also seems to be “fast art” for consumption, but that’s not the generator’s fault if the data is filled with art like this or if the majority demands it. However, I see it as a great tool for concept artists and testing out ideas.

I also don’t think “the emotion of the artist is lost” or “the suffering of the artist, the ‘essence’ of the art, is removed.” The trope of the suffering artist is rather disturbing, in my opinion. The question for me is more: who is the real artist? The prompter who spent hours tweaking the prompt? The image generator? The artists in the training data? For fun, I created a prompt for an Albrecht Dürer portrait as a cyberpunk hacker, which looked really good, but I would never consider myself the artist who created the image.⁹ I’ve seen quite a few people now posting AI-generated art as their own “digital art” with no manipulation or just a slight tonal shift in Photoshop. The problem is that we generally assume people who share art also made it, and it would be more honest to say, “I commissioned an AI to do this.” People also have a bias against AI if they know it was created by one.¹⁰

Which leads me to the exploitation of artists - and I would say, yes, there is exploitation, and I find the reactions rather creepy. This will change the job of illustrators dramatically, and they’ll be replaced by people who make good prompts? It sounds straight out of a cyberpunk novel, where the low-life artisan has a small advantage on the art market because they know the names of art genres or artists. There are already people offering their work for prices way below minimum wage.¹¹ It boggles my mind - no one would want to work below minimum wage, especially not in a field that takes skill and years of practice. The “adapt or die!” mentality that comes with this is really off-putting. Do we really expect artists to become prompt-fixers for minimum wage and release art into the world only to have it copied instantly? Honestly, I can understand if artists get reluctant to produce new art now. Even I, after spending so many hours with the generator, felt numb and confused - especially seeing how good it can produce photographs.¹² And then some people say, “No, don’t feel afraid, we need you. We need you to feed the machine!” It makes me shudder to think that an artist puts all their skills into their work, only to have people generate dozens of variants within minutes.

In the end, I think it’s all about authenticity - showing what is genuine and unique about ourselves. Which brings me to the debate: “Should you edit your pictures?” where AI is playing a bigger role here, too. The discussion is a bit weird to me, because people have different understandings of what “edit” means and why some want to edit as little as possible. There’s always manipulation in the analog photo process, depending on how rigidly you define it. The focal length of a lens manipulates the subject. The lab edits the results through chemicals or scanning. It’s ridiculous to assume an “authentic” picture or print exists. For some, an edit is dust removal or cropping; for others, it starts with heavy alterations. For me, editing is when I add things that weren’t there - like artificial clouds. I try to edit my pictures as little as possible. The kick for me is getting the picture I want with film, composition, filters, and lighting tricks - not with Lightroom algorithms that add dramatic clouds that never existed. But that’s just my taste; others use double negatives in the darkroom to copy/paste clouds, and I think that’s also creative and genuine. It’s just not my cup of tea. Too much alteration makes a picture feel overburdened and empty to me.

Some AI projects are really cool, though, and I would use them to some degree. For example, I had a Polaroid of a ferris wheel with dust I couldn’t remove. Using GIMP would have taken ages, but there’s an open source AI project that can remove all the dust.¹³ That looks promising, but sometimes I’m not sure if I want to use it. For the ferris wheel Polaroid, I actually like the effect the dust creates - it looks like stars at night, even though I took it in daylight. For other pictures, I’d gladly use it, and I don’t think dust removal makes a picture less “authentic.”

So what’s next? I guess we’ll see a reaction similar to what the impressionists faced with the invention of photography.¹⁴ “Authentic” art will play a bigger role, which also means not playing the rigged game and ignoring things and people that don’t inspire.¹⁵ That’s something I try to hold on to. I do photography for myself, not to impress the masses or follow trends. It doesn’t have to resonate with everyone. I try to go my own way, learn new things, and explore myself. If I find inspiration with an image generator, why not - as long as I keep being authentic.