“There are endless possibilities…”
The message is blinking in front of me on a Discord server. An invitation to let my imagination go wild and start creating – to produce my first piece of AI Art.
Where to begin? A Studio Ghibli-esque melancholic whale in a tiny fishbowl? A painting of a tiger performing on a trapeze? A life-like photo of an older man taking a research survey?
These were just a few of the prompts I tried out on the popular AI Art platform Midjourney (accessed through Discord) – and the results it spat out were, I admit, pretty impressive.
But the world of AI Art isn’t all impressionist astronauts and castles in a 90s cartoon style (some of my other prompts) – it’s a space that’s as fascinating as it is worrying, raising a host of functional and ethical questions that will be reverberating through the creative industries for many years to come.
What is AI Art?
Over the past year, AI Art has been slowly seeping its way into the mainstream: dominating our social media feeds with ‘Magic Avatars’ that illustrate users’ profile pictures; providing illustrations in magazine articles; and being used to create children’s books. A piece of AI Art even took first prize at last year’s Colorado State Fair in the US.
But what exactly is AI Art?
AI Art is computer assisted art creation, in which artificial intelligence scrapes the vast catalogue of the internet in order to ‘learn’ an aesthetic, and then uses that learning to create new images in response to a user’s written prompt.
The artistic results of these prompts can be powerful and emotional… but they can also be downright weird and horrific (take a look below at my trial of a ‘cute’ picture of an animated dog with a fish in its mouth… *shudder*).
There are a range of AI platforms that people can use to create this art: OpenAI’s DALL-E 2, MidJourney, and Stable Diffusion are some of the leading players here – as well as a range of third-party apps that are built on these systems.
What’s the controversy around AI Art?
While it can be fun to play around with these tools, AI Art is not without its critics.
A flurry of op-eds have been penned in the past few months condemning these systems – and designers and illustrators recently staged a massive protest on art portfolio site ArtStation when they decided to display AI Art alongside human-created works.
So what’s got everyone up in arms?
One of the biggest controversies in the AI Art space surrounds attribution. When AI systems scrape the internet for sources, they do so indiscriminately – regularly pulling in works from artists across the web without their awareness or consent.
The founder of Midjourney has stated publicly that their system does not consider copyright when scraping for source material, and certain Lensa users have noticed distorted signatures and watermarks (presumably from artists of the original works being used) appearing in the app’s Magic Avatars – reinforcing the idea that this is a wider industry problem.
Augmenting this is a broader worry that AI Art will soon be used to replace paid illustrators and designers – a horrific irony, wherein the tools trained on the hard-work of human artists (without compensation) might one day be used supplant them from their jobs.
It sounds futuristic – but a recent article from the Bulwark credits Midjourney exclusively for its header illustration, there’s a puzzle game on Steam created entirely from AI assets, and cartoonist Carson Grubaugh is producing a new comic book sourced completely from AI image generators. It turns out, the future may not be that far away after all.
Beyond attribution, there are also some important questions about the source of AI Art (ie the places it ‘learns’ from) and what that means for the works being created.
As anyone who has spent long enough online will tell you, the internet is not a sanitised space – it’s full of awful, dark corners populated by people creating pictures that can be upsetting to many. When art is being scraped indiscriminately from across the internet, including from these less salubrious sources, should we be worried that harmful stereotypes and biases are being baked into the results?
Take for instance claims of the hyper-sexualisation of female avatars on Lensa – where one user found that of the 100 profiles she generated through the app, 16 were topless and 14 were dressed in skimpy clothing (almost one in three altogether). AI systems can struggle with racial biases as well: early reviews of DALL-E 2 noted that images of people tended to reinforce racial stereotypes – for instance, Wired reports that 8 out of 8 attempts at rendering ‘a man sitting in a prison cell’ or ‘a photo of an angry man’ returned images of people of colour.
AI Art Moving Forward into 2023 and Beyond
So with all of these controversies in mind, where does AI Art go from here? The answer (as with many topics in the emergent tech space) is: ‘it’s complicated’.
Proponents of AI Art see it as something akin to collage or the use of reference photos, and imagine a future where these platforms become just another tool in the art-creation toolkit – providing a strong foundation on which art can be developed and tweaked by professional artists.
Others (such as Midjourney’s CEO) claim that these systems provide an entry-level way for beginner artists to refine their craft – and even give non-artists the ability to create in a way that they couldn’t previously.
Some in these camps imagine a profit sharing model akin to something like Spotify – where creators who have been scraped by AI tools receive some form of compensation and credit for their work.
Legally, this utopian picture of AI Art’s future lies in a bit of a grey area – with lawyers on both sides of the debate arguing hotly about copyright and commercial usage.
The Verge has a fascinating article on this if you’re interested, and points to some tests (such as this case against Midjourney, Stability AI, and DeviantArt) coming down the legal pipeline in the States that may start answering some of these key questions.
With legal questions hanging in the air around this artform some platforms – such as Getty – are placing bans on users uploading AI Art. And art that even *looks like* it was computer assisted is facing some pushback from communities – an artist on Reddit recently had a piece banned because the subreddit moderator felt it looked too similar to works churned out via AI (even though the artist claimed otherwise).
To wrap it all up, it’s clear that debates around AI Art will be with us for the foreseeable future. It’s definitely a rich space when considering media products, and one that we at SketchBook will be paying close attention to going forward – particularly as AI-generated video starts to hit the headlines in the next few years.
For my own part - and bear in mind I called my company SketchBook - I very much come down on the side of the artists in this debate. It’s hard to see the benefit for AI Art beyond providing a route to avoid paying creators. While the owners of the big AI platforms will talk about machines learning and growing by viewing other artists’ works, I believe this is plagiarism.
We should not forget that while these works feel like they’re being created quickly at the press of a button or from a quick prompt dropped into a Discord server, they’re in truth built on the years and years of experience from real artists who are receiving no compensation or credit for their contribution, which doesn’t feel like something we should be celebrating in 2023…
…Even if does allow me to finally create the perfect picture of Pikachu driving a classic sportscar on the Champs Elysée.
While you’re down here…
Thanks for reading!
If you enjoyed this, perhaps you’d like to also sign up for the SketchBook Newsletter and get thoughts like these delivered straight to your inbox? Or you can always follow us on LinkedIn or Twitter.
And if you have questions about any of these topics (or some research you’d like a hand with) we’d love to have a chat - just drop us a line!