Who's afraid of the big bad code?

An Essay in defense of AI art

During the last six months (give or take) since Midjourney and similar AI diffusion models hit the mainstream there has been uncomfortable conversations in coffee shops and lunchrooms regarding their quality, speed and the future. But the quality of what these AI’s could produce was seen as mostly amusing and only rarely threatening. 

But with the advent of new algorithms, like Midjourneys “V4” - and by extension, the threat of the inevitable “V5”. A sudden roar of frustration has hit websites like Twitter and Artstation. Artists (many of whom I respect and whose opinions I hold in very high regard) have seemingly lost their heads. The dialogue has shifted from concerns (some valid) about Artstations ability to act as a platform for hosting portfolios. To a conversation about theft.

This new topspin on the conversation strikes me, genuinely, as wrong. I will  (as briefly as I can) address the more legitimate concern of “authentic portfolios” before diving into the discussion of our digital Keyser Söze.

But before we venture onwards there are two things that feel important to address:

  1. I am not affiliated with Midjourney or any similar AI company - though I am a fan of the technology and its future. I will use Midjourney as a proxy for the industry at large as it is the one I (and most people on Artstation) are most familiar with. 

  2. Let this be a fair warning to those for whom this discussion is already a settled matter and a sore point - “here be dragons”.

Also, if you want “ammunition” against my point of view, here’s an excellent blog post by one of my favorite MTG illustrators, Karla Ortiza, highlighting some very interesting points and concerns.

As a 3D modeling teacher at a vocational school in Sweden, it is important to me that my students are able to demonstrate their skills and progress on Artstation in order to increase their chances of obtaining internships and eventually being hired. So the issue of authenticity on these platforms has been a topic of considerable weight for me. However, at first I was concerned about the impact of AI-generated 3D renders on people's perceptions of what a junior artist can reasonably produce. However, in the world of 3D modeling, the process, breakdown, and technical aspects (such as UVs, textures, and edgeflow) are just as important, if not more so, in the hiring process. And to my knowledge, AI is not able to produce these things in addition to a cool image of a 3D render. These technical showcases/breakdowns demonstrate a thorough understanding of the software and techniques used, and are highly valued by potential employers. In fact, they may even be more important than the overall shape or appearance of the model. And for 2D illustrations, a similar breakdown is often both appealing for an audience and relevant for those looking to hire. Sharing the “making of” will cement your work as authentically yours and even give you the opportunity to look god-like as your fast-forwarded process video seemingly produces gold out of thin air!

If you work in other artistic fields, as a one-off illustrator creating cover art for fantasy novels, album covers, or meme’y and edgy t-shirt prints, or similar "single purpose" items, then in as kind words as I am able - maybe you should feel threatened by these AI algorithms. But only if what you are afraid of is being replaced by someone/thing that creates generic, stiff, boring, samey, BUT technically sounds illustrations. Because at least currently, that's just about all these AI are able to produce - and more-over they're not even able to reproduce their results or actually output what was asked for if the prompt ever goes beyond "Darth Vader with a top hat". When browsing the Midjourney community feed and comparing the AI output to the prompt that was used to create it - it’s clear that the AI doesn’t actually “understand '' what we’re telling it. Prompts are more of a suggestion than an order and the more of a story you attempt to have it make - the less reliable the outcome seems to get. Being able to produce the thing we tried to make (or was contracted to) is what makes someone a good artist - so with this metric every current AI is fast, yes, but very “schizophrenic”. 

I think that what makes someone a top level creator is the ability to create something new - no AI will be able to do this if its frame of reference is only things that already exist - which will be an insurmountable pinnacle AI will never reach. “No Lovecraft no Cthulhu”, “No Cthulhu no cosmic horror”, “no cosmic horror no reference points for you or me” (or the AI). “No tolkien no anything” almost - all D&D campaigns ever are basically little more than thinly veiled “I can’t believe it’s not middle earth”. 


The actual list of professions who are (really) threatened by the advent of AI image generators is vanishingly small and are in my opinion maybe those that we could frankly learn to live without. And this is me fully buying into the doomsday vibe.

Anytime a company, project or contract asks an artist to produce multiple pieces (of anything) that follow a very clear set of specifications and has to be self-similar between them. Then an AI is nothing more than a jumping off-point. A way of shaking off the metaphorical dust and getting some initial inspiration or maybe parts of a generated image can be lifted out and repurposed - like how Photobashing is already used in the same way. No AI actually knows how to repeat a character in different poses, ensuring that clothes, hair, colors and such stay consistent.

To me it’s no surprise that Artstation, a platform dedicating about 60% of its front-page bandwidth to the same generic, big chested, anime-ish cheesecake character concept illustrations would be the first to lose their collective marbles. Nothing surprises me more than those who spend tireless hours just to create the same vapid softcore. So if AI can eventually create those same stiff mannequins then not much of any real value was lost in my eyes. “Character design” should be about using design to showcase someone’s mental state, profession, philosophy and story (you know… their “character”) - but I would like to stress that “being hot” is not a personality trait unless you’re the Human Torch. And at least the most prominent AI image generator, Midjourney, doesn’t allow for nudity, so one could finally go “all the way” and have a watertight reason for doing so - win-win!

To summarize my position, if your ambitions stop at creating generic, shallow art, then don't be surprised when a mindless generic and shallow AI is able to eclipse you. But if your ambitions go beyond that and you strive for creativity, depth, and the ability to tell a meaningful story, then you can rest easy knowing that things like being specific, dynamic, consistent, and able to tell a story are still a long way off for the AI (and maybe so out of reach that it will never manage it).

AI is not leaving. But maybe it’s leaving Artstation we’ll see if the mob tires before the platform caves. In my opinion I don’t feel like attempting to remove AI art is the appropriate way forward - and to be fair, I’ve seen some of those who want AI regulated on the platform echo these same points. The most reasonable suggestion seems to be to allow users to themselves tag their works as being AI generated or having used AI in the process. And for those who feel that their integrity and talent should not be fair game for the AI, a way of blacklisting your artwork could possibly be solved between artstation and the popular AI companies. Though for 99.99% of users on Artstation, you’re not (by definition) the top 1% and should almost be flattered if someone wants to have an AI (or anyone) take inspiration from your work. And if you are in the 1%, you will land on your feet.

As of writing this essay, Artstation has actually gone ahead and implemented both of these suggestions - you would think that this would give the platform some good will in the eyes of its users - but still the mob has not died down. Making my theory of people simply being miffed that any teenager with an internet connection can do in minutes what takes them hours far more reasonable. “The war continues”, “round 4”, they claim as they charge towards their next windmill. Here’s a hot tip, if you are not satisfied, despite the blazingly fast response from Artstation - make your own portfolio website. And I don’t think I could even imagine a better advertisement for Midjourney than (at least the perceived) majority of the largest artist website in uproar over the increase in quality of the output of their latest software update. 

The fact that an AI is able to produce more mainstream appealing images than I am - and faster to boot. Is in my eyes no different than me not being able to make what someone like Frank Franzetta was making. I am quite happy with where I am in my art “journey” and the presence of faster, younger, and more skilled artists has never stopped me from immensely enjoying what I’m making. We all started out creating what we would probably call garbage (or hopefully novice) artwork today, we knew that others were better than us then but this had very little impact on our quest for progress, or the enjoyment we got out of drawing. So what I’m saying is that even if someone who never drew before could whip up some jaw-dropping AI art it would be foolish of me if I got frustrated.

What does make me feel frustrated is that the conversation has headed in the direction that it has, I see (at least) three very exciting prospects related to using AI in art. The conversation seems exclusively focused around what we’re losing instead of what could be gained.

The democratization of generic quality. Indie-developers, card game designers, self published novelists and roleplaying enthusiasts now all have means to create illustrations to help bring their visions to life. 

Idea generation and bulk production is tiring, time consuming and pretty uninspired. Creating those silhouette figures scratching seemingly in random in order to eventually develop a blob into something exciting. This initial step could easily be given to the AI without anything meaningful being lost in the development of the final product. Time to go to lunch, why not ask the computer to keep itself busy churning out some 100 odd blobs for you - the art director and the one with the cool ideas and taste for good and bad to then pick and mix from.

Less grunt work seems to me to be a real positive, when it comes to budgets, deadlines and revisions having to take hours and hours to create a “who-cares-just-make-it” figure or asset is not something that many senior artist would care if it got off-loaded onto the intern. Let's use AI to help us find the time and money to make those assets and key visuals that we really enjoy making and leave the boring stuff to automation.  “Get me pictures of spiderman!”

I envision a future where anyone with a strong vision will have the time, tools and the resources to help them create something that is, to them, spectacular. 

So finally, in regards to the topic of thievery. What exactly was stolen? I implore all of you who have stuck with me (or scrolled) this far to specify in no unclear terms what exactly was stolen - because to me, all that you had and were still remains. No artwork was removed from the metaphorical walls of your personal museum and no skill or proficiency was sucked out of you by some AI-dementor. Theft implies that someone lost something. 

The way you (and I) copied, borrowed and was inspired by our favorite artists, creating amalgamations of different influences by taking “an eye here and a pose there” is to me no different from what the AI is doing, only the scale is different.
Karla Ortiza makes some compelling points in favor of there being a distinct difference between the human and AI way of “learning” - I am somewhat sympathetic to those arguments but I feel like they give to much credit to the imagined purity of the human, Ortiza claims that an AI can only produce images based on what it has been trained on, but this is only sort of true. It is readily apparent that these AI models are more than able to combine two or more concepts in novel ways, and in the same fashion I believe that this in large parts is exactly what we humans do - and it misses the point of there being a co-creation aspect attached to the prompt writing process.


There is a very relevant discussion to be had on the topic of overfitting, something that Midjourney’s V4 seems to be more prone to in specific cases than its previous iterations - but I dont think this extends past a select few world famous paintings. But should there surface proof that any AI algorithm overfits to a level that would raise some red flags I’m of course in favor of those being taken down or drastically modified. 

I do believe that you consented to having someone like me draw inspiration from your artwork and mimic your style the moment you uploaded your stuff to the internet, not unlike how we can,  very clearly, see influences of Bill Sienkiewicz in the works of Dave McKean - something that I for the record, find absolutely nothing wrong with - this process of borrowing (sometimes heavily) is part and parcel in all creative fields. We all stand on the shoulders of giants. I also believe that this process of finding something really cool and wanting to make something similar is the driving force behind much of the motivation for artists - but to call it a wish to “become someone else” is somehow frowned upon. I find no difference between how I might try and mime your style and how the AI has been trained to do the same.

I feel a looming sense of entitlement coming from some artists who have (and I admire this) spent a long time developing their craft. To see an AI be able to create images that an audience finds as cool or as appealing as what you made can be disheartening at first. But this admiration or recognition was never yours to take - but for others to give, and if an AI raises the bar of quality then you can either sink or swim.

I admit that the speed of the Midjourney algorithm in relation to the quality it outputs is pretty dizzying - but parallels to the “photography is not real art debate” which sprung up over 100 years ago feels to me very pertinent.

Did the AI take your source of income? Probably not, and in the case that it did it wasn’t actually the AI that stole it, but your client who decided that their money was better used elsewhere - or even more precisely, the steady march of progress took your job. For sure I have a lot of sympathy for those suddenly finding themselves without their usual revenue streams - but automation has changed the landscape of a tremendous amount of jobs, and there is no intrinsic value or sanctity in what you do. Working in a factory and getting paid to make an illustration are just two jobs. 

If someone asked me for a good word to describe a method in which an artist drew inspiration from multiple different sources, took the most appealing bits, drew new versions of all of them and then assembled them into something new - I would be hard pressed to find a better word for that process other than “creative”.

The only thing the AI might have stolen from you is your sense of being special. I get it, that’s blunt, but as un-nuanced as the reaction has been on Artstation this harshness is in my eyes well deserved.

We’re in total agreement that AI and its impact on different fields and professions should be monitored and in cases where it’s malicious it should not be tolerated. I simply disagree that this is what is going on.

“Good artists copy; great artists steal.”

– Pablo Picasso


With reservation and love 

Max

Max FribergComment