Throughout 2023 there was a load of hooha around AI (artificial intelligence) in the news and online. Many writers, artists and makers seemed to be lining up to say how awful it was, but I felt pretty open-minded about it. I was persuaded by the opinions of people like Joanna Penn from the Creative Penn podcast, who basically said she viewed it as a tool that could work as an “amplifier of human imagination” (as opposed to a replacement for it), but that as with any tool – e.g. the printing press, cameras, the internet – that tool could be used by people in either negative or positive ways.
I was particularly interested in the mention of text-to-image generators, because it appealed to all of the different versions of ‘me’: the me with an MSc in computing, the me with an MA in creative and critical writing, the me with an accidental MRes in education, and the overall me who loves learning and making new things.
So at the very end of 2023 when my Christmas craft rush was over, I thought I would have a play with some AI image generators to see if they might be useful in my own artwork…
…and it was FUN!
This is the first ever AI-generated image that I produced, using Adobe Firefly: [oh by the way, the links on this page are all just for information – none of them are affiliate links and I’m not sponsored by any products]
I was completely blown away. I’d used the prompt “Letter ‘K’ with doodle flower art, output as a vector image for a papercutting design“, and it came out looking like one of my own sketchbook doodles – only better.
However, I discovered that even though Firefly has a specific theme of ‘Vector look’, I couldn’t find a way to get it to actually generate SVG/vector images. Maybe this will change in future, but at the moment (January 2024) it looks like Firefly can’t generate vectors so you have to use Adobe Illustrator instead. I did start a free trial of Illustrator, but (1) it was really stingy with the number of images that could be generated, (2) the images were so rubbish that I never even bothered saving any, let alone using them, and (3) considering those first two points, I couldn’t justify going ahead with a paid subscription when the package didn’t do what I needed it to do.
Having tested Firefly and Illustrator, I thought I’d try out another AI image generator. After a Google search trying to find an image generator that could output vector graphics, I discovered Kittl. Like Firefly, it produced some really interesting initial results for me…
…but like Firefly, it had some drawbacks. Again, everything may have changed by the time you read this, but at the moment Kittl seems to be unable to reliably create anything with letters or words in it.
For example, can you guess what letters these are below? “X”? “B”?:
Nope – they’re both the letter K, according to Kittl. Sigh. Back to the drawing board (literally).
So I went back to Firefly to see if it could handle letters any better than Kittl. It could (mostly). Here’s what I got with the prompt “Large serif letter ‘J’ decorated with waves” I really liked it – even though the decoration looks more like a feather than waves:
I saved the image, loaded it into some software that can convert images to SVG files (“Curve“), and then tweaked the image a bit.
And so I created my first-ever piece of AI-assisted laser art. Because after editing parts of the SVG file, I etched the J onto a “scribbles that matter” bullet journal for my sister, and gave it to her for Christmas.
Instead of taking hours to create, it took 10 seconds to generate the initial image, then about half an hour of tweaking the vector file. I am now thinking that AI will supercharge my creativity and artistic output, because…
AI-assisted designs versus my traditional method
Let’s compare the AI-assisted process to my usual method of creating a design:
With my traditional method, first I round up lots of reference images (from books, my own photos, and/or other media like Creative Commons images, licenced images, and images that are in the public domain.) With all of these images I then begin to ‘synthesise’ elements and shapes in my sketchbook – i.e. combine them into what feels to me like a summary of the images.
[IOU a screenshot of my digital sketchbook here, but in the meantime just imagine a load of seagulls in different positions and from different angles]
The eventual design is therefore an original artwork that references all of those other images, but doesn’t copy them. What I’m aiming for is to try to get to the concentrated “essence” of a shape. So from the images above I am attempting to create the most seagull-y seagull possible. I want people to look at the image and instantly think, “Yep, that’s a seagull”. It should be a design that represents a seagull. It’s not one specific seagull that really exists in the world, but rather an amalgamation of many examples which make up the seagully-est design I can manage.
Example of the final seagull on a fridge magnet, which I’ve been selling via some small shops and galleries:
The whole process often takes between two and five days (I tend to take longer for historic buildings than I do for animals). This is fine for products that I will be able to sell in large quantities, because the cost of my time will eventually be repaid – albeit in very small chunks – e.g. a fraction per sale from each fridge magnet or Christmas bauble. And that time is also OK for pieces that I make for family or friends, because I’m creating the items purely for their (and my) enjoyment.
But what happens if someone wants to commission a piece of unique artwork from me? If it takes me a day to design and make, then I’ll probably have to charge over £100, which would put my work out of reach of most people. Plus I’m also not currently well-known enough as an artist to attract many of those customers anyway.
The way that papercutting artists usually get round this pricing problem is to spend time creating a template, which has the same overall design but key parts of it can be customised. For example, my Christmas baubles share the same filigree designs, but can then be adapted to include individual names inside them:
What I would really love is to create a completely unique item for each person, but for the process to only take me about an hour instead of a couple of days…
…That’s where the AI comes in.
Remember the “artistic synthesis” stage above, when I created the most seagull-y seagull that I could manage but it took me two or three days? This is where AI might be able to revolutionise my design process.
AI engines are basically able to follow the same process as me, but in about ten seconds instead of ten hours. They can access thousands of images of seagulls, compare them to each other, and come up with their own version of a seagull based on the similarities in the images. Great!
Unfortunately, the AI doesn’t actually know what it’s doing. If it produces an image that is pleasing to humans, that’s probably pretty much by accident. It may have come to a similar output design as mine, but it’s done it through brute force and ignorance rather than by artistic judgement. Probably the only reason it has come up with a human-pleasing design is that humans tend to only upload/share images that they actually find attractive in the first place, so those are the inputs on which the AI is basing its outputs.
My theory is based on experiences during the last couple of weeks. I’ve learned that AI image generators are capable of producing some nice-looking images, but that they make errors that no human would ever make. For example, the images below show that the Kittl AI came up with some recognisable designs of ballerinas…but they featured anything between 0 and 4 legs. I would suggest that four out of those five options represent a non-standard number of legs:
And since when do ballerinas wear stilletos?! Similarly, in the last two weeks I’ve seen an awful lot of 2-tailed cats, plus a 14-fingered woman (6 on one hand and 8 on the other, just for info).
Yes, but is it really ART?
This is not a new debate. Marcel Duchamp famously sparked outrage with his 1917 “sculpture” piece called Fountain. (It wasn’t a fountain – it was a mass-produced urinal.) And heck, is photography really art? Over 100 years after these questions originally surfaced we are still basically asking the same questions about art and/or artists. Can Andy Warhol be considered the creator of all of his Factory of artworks? (check out Tim Harford’s Cautionary Tales podcast episode for a good discussion about Warhol’s work.)
I think for most people there is a sliding scale of what counts as an “artwork” or an “artist”, and what counts as “original”, “influenced by”, “derivative”, “pastiche”, or “blatantly plagiarised”. It’s often a very personal decision, depending partly on the intent and skill of the artist, and partly on the interpretation of the beholder. Billions of digital photographs are taken every year, but surely only a tiny percentage of those photos are intended/interpreted as art, even though they may have all been created using the same basic equipment.
AI generated art versus AI assisted art
I personally am convinced that the images that AIs come up with are at least original. Every human artist has been influenced by other artists or artistic movements – whether consciously or not – so as long as the elements of the AI reference images have been legally obtained and have genuinely been influenced rather than directly copied then the method of creation seems OK. If it’s good enough for humans, it’s good enough for AI.
But for me, there is a difference between AI generated pieces and AI assisted pieces. Thanks again to Joanna Penn of the Creative Penn podcast for bringing up this distinction. AI can produce some fantastic images, based on prompts by humans. Do we call those humans “artists”, though? At the moment I’m leaning toward saying yes, because I can think of established artists who already use pre-existing objects in their artworks. One of my favourite artists is Andy Goldsworthy, whose work has included ice, leaves, rocks, branches, twigs. When I think about the processes that he must go through in order to get from the start to the finish of an artwork, I could argue that they are vaguely similar processes to those of people who work with text-to-image generators. They have to come up with an idea or prompt in the first place, but then (if my experiments so far are anything to go by) the artists may have to discard tens or even hundreds of the generated images in order to end up with just the one “right” image. I’d bet that Andy Goldsworthy doesn’t use every leaf or rock or ice shard that he sees while he’s creating his artworks, and that he carefully selects exactly the right objects for exactly the right positions. And I can’t remember who originally declared it, but don’t many photographers say that the secret of being a good photographer is to discard all of the bad photos? The perceived artistry therefore comes partly through the ideas and decisions that the humans have made during the production and/or presentation of their works.
That makes a lot of sense to me, so I’m satisfied that people who use unedited AI-generated art do have some claim to being creators of art.
However with my own work I’d like to go one or two steps further than just churning out images that have been made for me by a computer. For a start, I rarely get a result that looks perfect to me. Nearly all of them have some element that either looks plain wrong (three-legged ballerina, anyone?!), or maybe doesn’t fit well with the rest of the image, or doesn’t match the prompt words. Therefore I almost always have to change the prompts, and to find ways to edit the images to make them fit my notion of “good”.
After that I’ll convert the image into an SVG file and then edit that file for another hour or two. Finally, I’d like to actually turn the image into some kind of physical item (e.g. like the “J” bullet journal above). Maybe more about those additional stages in a later post, but for now I’m just happy to report that because of the stages my work goes through, I regard my recent experiments as being AI assisted rather than AI generated. An awful lot of me will be going into the final pieces, and the main difference between these artworks and my usual ones is that a lot of the time-consuming “synthesis” stage has been speeded up for me by the use of an AI tool. I feel like it’s the equivalent of Andy Goldsworthy taking a team of assistants with him when he’s going to make a leaf-based art installation, and saying to each of them “Bring me a bucket of orange and yellow autumn leaves that don’t have any bits of green in them”. He would still be creating the actual artwork, but would also be saving himself some time. I guess it’s also like painters buying their paints from shops, instead of grinding and mixing the colours themselves.
This is only the very beginning of my adventures with AI-generated artwork, and I’m hoping to explore them a lot more during the rest of 2024. I know it’s still a very contentious issue with a lot of people, but for me, for now, it’s FUN. I think that as long as I’m honest about my creative processes then it should be up to me to use whatever tools I want. Just as I wouldn’t try to pass off my laser-cut artworks as being cut by hand, I won’t be telling people that my pieces aren’t influenced by an AI if they actually are. On the other hand though, if I’ve spent a lifetime enjoying Art Nouveau and Celtic knotwork, I don’t think I should be expected to calculate what percentage of each of my artworks has been influenced by those genres if my pieces somehow all end up with flowers or sinuous lines in them.
Creative goals for 2024:
Bearing all of the above in mind, I really do want to continue exploring AI text-to-image generators, and to see where they might lead my work. Therefore here is a ‘goal declaration’ for 2024: