The AI art revolution, pt.2: Benefits

In part 1 of this article, I’ve detailed the natures of AI image generators, art, and artists, and laid out some of the issues with the technology. Part 2 is about where we go from here. The tech exists now: What can we do with it? I read many tech-bros suggesting AI image generators as a super useful tool that artists should embrace, but they tend to be unfamiliar with the processes of making artwork. Having spent a decade in artistic communities, I can think of several applications that I and other artists could probably get behind, if the moral issues were settled.

Artist’s block:
Brainstorming with AI imagery is the most suggested yet least required application, but I will specifically note the “artists’s block”. Just like writers, artists sometimes run out of ideas or motivation to draw the same old, which can be a real obstacle to one’s productivity and consequential sense of purpose. The range and unpredictability of image generators could be of particular benefit here: Artists could type in rough ideas, or use some service that is a combination of word salad and image generator, and seek inspiration in the randomness of the machine. It may serve better than browsing a portfolio site of perfected artworks that leave little to the imagination.

More ideas than time:
On the flipside, creatives tend to get new ideas faster than they can put them to canvas. Not every idea is of equal importance however, some are just cool, conceptual, or contemporary social commentary, and don’t require the same full-time dedication as a masterpiece. Instead of leaving conceptual sketches at the bottom of a pile, one could put these ideas to the machine and have a shareable coloured image portraying the gist of it, satisfying enough as a momentary outlet.

Concept pitching:
The beginning of a commissioned artwork is always to figure out what the client wants, which may not convey well in words. To guide the client, an artist typically makes thumbnails and sketches of different ideas to choose from. The back-and-forth in this early stage of design can be frustrating when clients request revisions upon revisions without committing. The creation of these “sketches” could be delegated to AI image generators, who are already set up to generate multiple crude versions from a description. It may not take much less time, but it may be less taxing, as it is not the artist’s personal creations that the client is shooting down.

Reference images:
Artists often use photos as references for subjects that they can not draw entirely by heart, such as vehicles and horses. Since Google Image Search removed the option to download images directly from search results in 2018, there is demand for new sources. AI-generated reference images could in theory be a step up, as their results are not limited to existing images with regards to pose and perspective. In practice however, AI image generators in 2023 are too slow and erratic for the demands of artists, taking a minute to produce hands with seventeen fingers. Perhaps this will become more viable in the future.

Most artists are skilled at drawing people, but less practised at backgrounds. This is partly due to the art world’s preoccupation with the human form, and partly due to the greater diversity of environment objects. In many cases, a background’s role is only to give a general impression of the surroundings, to set a mood, or just to fill the void. Drawing detailed buildings and vehicles is an expertise of its own that can take as much time as drawing characters. I think many character and comic artists would welcome an option to auto-generate backgrounds, so they can focus on the humanity in the artwork.

Art theft:
Though of questionable nature itself, AI art may reduce traditional art theft by a lot. Would-be artists with more time than skill are already getting the attention and admiration they want by publishing AI arts wholesale, without having to impersonate other artists. Event organisers that previously repurposed downloaded images to promote local dance parties, can now have something equally ill fitting AI-generated for them. Bootleg sales of e.g. T-shirts prints may in part be replaced with easier to acquire AI arts, but probably less than other forms of art theft, as existing arts still come with assurance of popularity.

Low budget clients:
A common plague upon artists are clients who demand masterpieces for the ridiculous price of $15, and will not take no for an answer. Rather than get into an argument over the value of educated skill, artists can now give such clients exactly their budget’s worth: An AI-generated image of Spyderman, instead of a hand-drawn Spoderman. I expect this tier of clients to continue to exist, because not everyone will want to sign up or pay for a good AI image generator.

This one may be controversial, but there is one task in the industry that I consider inhuman: ‘Tweening, i.e. drawing each frame of animation be-tween key frames. Hand-drawing 30 images for a fleeting second of animation is an arduous and costly task, and that is why cartoons use a less fluent 12 frames per second, have simplified character designs, or 3D animation. Due budget restrains, tweening is often outsourced to Japan or Korea where animators work 12 hours a day, sleep in the office, and build up Repetitive Strain Injury. Using a mix of image generating AI and other existing algorithms, it is possible to delegate more than half the animation frames to machines, enabling higher framerates, reducing labour intensity, and potentially upgrading grunt-work artists to lead animators. After mass layoffs, though…

Adoption of AI art generators so far

None of the dozens of artists I know personally have started using AI image generators in their work. There are stories of artists whose superiors decided to switch to AI, whose jobs have now lost meaning to them as the creative aspects vanished. There are well established artists who have taken to using AI to increase their output of NFT arts (digital images that you pay to have ascribed to your name) , but it seems to me that the selling point of NFTs’ uniqueness conflicts with the infinite variations of AI-generated arts. There is a small percentage of artists that is happy to use AI image generators for brainstorming, but most established artists are still strongly opposed to using AI image generators, not in the least because using their personal art to train AI without asking was a bad way to start any kind of collaboration. Let no-one be fooled: AI art generators were not meant to be tools for artists, no artist asked for them. They were made just because it could be done, and it’s up to us to decide how to use them.

Fervent adopters of AI art instead seem to be newborn enthusiasts flooding art sites with such unique works as “Untitled”, “Untitledsdfsdf”, and “2349423037710”, whilst wielding more fees and Patreon subscriptions than other artists. It seems AI art has become the next get-rich-quick scheme, and given the amount of views and followers AI art gets, I’m not sure they are not temporarily successful.

Art portfolio sites have implemented AI art as a category of its own, but their recommendation algorithms still favour them by virtue of sheer quantity. Some art sites have chosen to credit the name of the algorithm instead of the name of its user, which seems reasonable, given which has the most influence over replicable results. One well-known portfolio site, Deviantart, introduced an AI art generator of their own in 2022, trained on all existing artworks of its users without asking for permission first. After mass outrage, they changed the default setting to being excluded from their AI’s training dataset, and introduced a technical tag, “noai”, as a measure to discourage AI art scraping from Deviantart. Although a few other portfolio sites have adopted their “noai” tag, it has no actual effect, even according to Deviantart’s own terms of service:

“Users acknowledge that by uploading Content to DeviantArt, third-parties may scrape or otherwise use their works without permission. DeviantArt provides no guarantees that third parties will not include certain Content in external data sources”

So far, print-on-demand merchandise sites like Redbubble and Teepublic seem to be spared from flooding because they require high-resolution images, but it won’t be long before AI art enthusiasts tie image generation to design submission, as the most successful sales tactic on merch sites is to have hundreds of designs. Soon after, I expect the servers of print-on-demand services to collapse under the quantity of user design submissions, forcing them to drop user submissions entirely, and replacing them with AI merch generation directly to consumers, as other parasitic sites already do.

Overall, we are seeing large increases in mediocre contents and little to no meaningful adoption by established artists. I would like to see AI image generators used in the areas I mentioned, but the way forward is still blocked because the AI art revolution started off on the wrong foot: Exploiting the work of existing artists instead of supporting it. Until justice is done to that initial foul, artists will not collectively adopt AI art. Even if some artists wouldn’t mind delegating part of their workload to AI, the art world is a community built on mutual admiration and support, and using other artists’ work without permission, even indirectly through AI, does not sit well with them.

So, in what will be part 3 of this exceedingly long article, I will finally address copyrights.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s