Some thoughts on AI art tools

AI generated images of roses

First three results from Craiyon using the prompt a haunting of roses

Publicly-accessible AI art tools — Craiyon (formerly DALL-E mini), Midjourney, and the growing number of others — allow non-skilled users to produce an image that otherwise they would not have been able to create themselves. This is done by the software amalgamating visual images made by human artists, feeding them through various algorithms, and applying filters to produce a blended image. It’s all rather clever.

At the moment many shared images created with AI art tools do have the feeling of a selfie that has had far too many phone/Instagram filters applied to it to hide blemishes and imperfections - and possibly for the same reasons. This is something that will probably change very rapidly with technological development of the AI art tools, though.

Obviously the software needs human-created images — paintings, drawings, photography, etc — as raw material.

Why? Because if, for example, your text prompt contains the word “rose” then the software initially needs a bank of human-created images of roses in order to create its own image of a rose.

Software doesn’t initially know what a rose looks like, but it can be given a massive store of human-created images enriched with metadata that contains the word “rose”, and with this it can then select and combine from those matching images to produce something that resembles — to us, the viewers — a rose.

More advanced AI art tools may even be able to use machine learning to identify new images of roses based on their existing store of images of roses (removing the need for metadata) — similar to the flash of text you may have seen on Instagram before the photo loads saying “May be an image of a flower”.

On derivative works and compensation for copyright owners

The obvious question that arises from using human-created images to create derivative work is whether the humans who created the original images will be credited or paid for the use of their work.

I’m presuming here that some, if not all, AI art tools are using a massive image database that may contain copyrighted images which the AI art tool does not have a licence to use. And please note that I’m using the phrase “derivative work” here in a legal sense, not in a pejorative sense.

The legal answer to this question is that any derivative work that is not substantially different to an original copyrighted work is copyright infringement (unless the original work is published under a licence that allows derivative works etc). The phrase “not substantially different to” is obviously up for extended legal squabbles.

The practical answer to this question under Western capitalism is that very few artists have the finances or legal clout to pursue any such claim, whether it has merit or not.

However just wait until one of the AI art tools catches the attention of a major corporate — Disney, for example — for the use of that corporation’s intellectual property and visual trademarks — characters or spacecraft from the Star Wars franchise, for example — in an AI-generated image, and things will change rapidly (but probably only for intellectual property and trademarks owned by major corporates who can sue the arses off the tech bros responsible for a particular AI art tool).

I suspect Disney et al are already keeping a very good watch on AI-generated art and are just deciding when and how to act. Or perhaps the developers of AI art tools have pre-emptively programmed their tools to avoid using artwork owned by any corporation rich enough to sue them - has anyone spotted Mickey Mouse or the Mandalorian appearing in any AI-generated artwork?

What the legal outcome will be is therefore unknown, but I doubt it would affect anything other than intellectual property and trademarks owned by big corporations (rather than being a blanket ban or compensation scheme for creators of any copyrighted works being used to create derivative AI-generated works).

Any compensation for individual artists whose works are used by AI art tools to create derivative works would probably need to be enforced at governmental or — more realistically for most countries other than the US — supranational union (e.g. European Union) level. If this ever happens it will be years or decades away, given the speed at which these large organisations move.

On the effect on the income of visual artists

There is a valid argument being made by some artists (particularly illustrators) that the use of AI art tools will take money away from artists: that people wanting cover art for products such as books and CDs, or people wanting artwork to illustrate articles in magazines or websites will use free AI-generated art rather than paying artists to create bespoke artwork or paying artists for a licence to use one of their existing pieces of artwork.

It is also completely understandable that using AI-generated cover artwork would seem especially ill-judged where those who would normally pay for the artwork are part of the wider arts community themselves - for example writers or musicians.

The complaint I’ve seen from some artists is twofold:

  1. not only does this threaten the income of artists/illustrators due to people using “free” artwork created by AI art tools, but also
  2. that the AI art tools may be utilising original artwork by those same artists to generate derivative AI-created artwork without payment to the artist (as copyright holder) for the use of the original artwork.

I don’t work as an illustrator and I don’t create artwork on commission, so the first point does not directly affect me. I have, however, had a handful of inquiries over the years from people asking to use existing artwork for CD cover art etc, but it’s definitely not a common occurrence.

For the second point it may be that — depending on where the AI art tools are sourcing their databases of artwork from — my artwork is being used to help generate AI artwork without any recompense to me for the use of my work, and without any licensing agreement to legally allow this.

The point of the two paragraphs above, particularly the second one, is to make it transparent that, as an artist whose work may possibly be used to help create derivative AI-generated artwork without my permission and without compensation, I do have a vested financial interest in this, and so I cannot be considered as a wholly objective and neutral observer. That said, I’ll try my best.

I think it is undeniable that at least some illustrative artwork (by which I mean any artwork that is used as cover art or to illustrate articles etc, whether it was commissioned and created for that purpose, or was existing artwork where a licence fee was paid to the artist to use it as such) will be replaced by AI-generated artwork. This means a loss of income to artists, particularly to illustrators.

One counter-argument is that the use of AI-generated artwork is just replacing the use of legally “free to use” artwork published by the artists under Creative Commons licences i.e. the people using the artwork would never have used non-free artwork anyway, so it is not replacing artwork where the artists would have been paid. This is certainly possible, but probably an untestable theory.

However if we do see “free” AI-generated artwork replacing the use of commissioned artwork from illustrators (or licensed use of existing artwork) then that is certainly a cause for great concern, especially in a cost-of-living crisis where freelancers such as illustrators are already on the frontline of cuts.

I doubt we’ll see any action from the current government as they’ve already shown numerous times that they don’t value the arts at all - and the same could be said of those companies or individuals who replace the use of paid-for commissioned or licensed artwork from artists with AI-generated artwork.

AI-generated artwork does have several downsides for use as illustrative artwork — the AI art tools often don’t generate it in high-enough resolution for print, a lot of it looks very similar in style, and you can’t specify exactly what you want depicting — but these are all technological constraints and the past few decades has taught us that technology advances very quickly. I wouldn’t be surprised if some or all of these issues have been solved within a few months.

As for point (2) above, I’ve covered that in the section “On derivative works and compensation for copyright owners” above, and I’m not optimistic for anything helpful to artists there for many years.

On the comparison with the invention of photography

There have been several comparisons made on social media between the initial public use of AI art tools and the invention of photography, related to its potential ramifications to visual arts.

The seriousness of these comparisons varies, and most of the comparisons are not qualified to any real extent. I want to use this section to examine this comparison in a bit more detail.

So, we can all now create images — utilising a vast database of images created by humans, combined with metadata and algorithms and filters — that we otherwise wouldn’t have been able to create, and very quickly too.

And in some ways this is indeed comparable to the invention of photography in how photography allowed anyone to create a “realistic” rendering of a real life object, person, or scene.

Of course, after a few years of photography being accessible to the public it became clear that some people could create much better photographs than other people for many reasons, but usually a combination of having learned technical skills with a camera (greater knowledge of focal lengths, apertures, film speeds etc), compositional skills, knowledge of lighting, not putting your thumb over the lens, and so on.

Will the same happen to AI art tools? Probably not at the moment, when the only input a user has is entering a text prompt (although perhaps there might already be some tricks in the selection of words or phrases in the text prompt that produce better results than others?).

In the future, if more user-facing controls get introduced to AI art tools, then it’s certainly conceivable that the experience and technical skills of a user will have a qualitative effect on the images they are able to produce.

The other thing that happened with the invention of photography was the acceleration of the other visual arts (painting, drawing, sculpture, etc) away from what we’ll call “representational art”, and towards other forms such as impressionism, expressionism, abstraction, and so on.

And so we can ask ourselves the question: will AI art tools affect the other visual arts (painting, drawing, sculpture, photography, digital art etc)?

If photography freed artists from the constraints of “realistic” representation because it was considered easier and cheaper with photography, then what does AI-generated art make easier and cheaper?

On the use of AI art tools as tools for artists

I can definitely see these tools being used by artists, initially for use in generating ideas and breaking artistic blocks by encouraging lateral thinking, much like the use of Oblique Strategies cards.

As they advance technologically, perhaps their uses might increase in this aspect.

On capitalists being capitalists

Come use our free tool! Love our free tool! Let our free tool become indispensable now that all the illustrators have disappeared! Yeah, now maybe it’s not free to use anymore…

We’ve seen that transition from “free beta” to “premium paid service” so many times in the tech bro business model that it’s become a cliché. But in order to pay the developers, the hosting & image processing fees (and the lawsuits from the likes of Disney!) then the makers of these AI art tools need an income.

They’ll run for free for a time, paid for by investors betting on the potential of a future profit, but eventually capitalism demands that they are monetised.

Initially that could be something like forcing users to watch adverts while the images are being generated (given that it takes a bit of time to generate the resulting images anyway).

Beyond that I can imagine intrusive watermarks being stamped across the images (unless you pay for the unwatermarked version), paid add-ons for new functionality or image styles, and so on.

This, in the end, might solve the problem of them replacing the use of paid artwork commissioned or licensed from artists, but how many working freelance illustrators will still be around by that time?

The only statement that is incontrovertible is that nothing is free under capitalism.

And finally

And I’ll (finally!) finish this blog post with one question to ponder: if the AI art tools are using artwork found online to create their derivative artwork, and everyone’s posting all their AI-generated art online, then how long before AI art tools start using AI-generated artwork as their original work?

What happens when AI art eats itself?