As a Software Engineer and (very amateur and very part-time) 3D/Digital Artist, I was naturally intrigued by some of the AI tools that are becoming available to artists. Then came all the backlash against AI art. At first, I wasn’t sure what all the fuss was about, so I decided to look into it (while trying to keep an open mind).
Will it lead to the death of the artist? Is AI Art actually… Art?
Here are my thoughts.
Introduction
Artificial intelligence (AI) has come a long way in recent years. Technically, ‘AI’ is often a misnomer. What we are usually talking about is known as ‘Machine Learning’. A Machine Learning system is simply a piece of software that has the ability to automatically learn and improve, based on data or experience, without being explicitly programmed. It isn’t any sort of advanced or sentient general-purpose Artificial Intelligence (so, don’t worry, we’re not even close to developing Skynet – or even Wall-E – at this point!).
I’ve used Machine Learning in some of the research projects I’ve worked on. These tools can be trained to do lots of really useful things like data and image analysis, which can assist researchers by performing large, complex tasks more accurately and in a fraction of the time. As well as well-known apps like ChatGPT, they’re increasingly being used in medicine, environmental science, language translation, and more.
But what about AI in art? Well, straight away, I think we need to establish some definitions. From what I can tell, there are two main categories of AI Art tools out there, which I will refer to as:
- AI Image Generators – which generate an entire image from a series of parameters and some training data.
- AI Assistant Tools – which perform tasks like generating textures or applying filters to digital models/images.
So, do these tools have their uses? Is there a reason to be concerned? Let’s examine them in more detail and look at some example case studies.
AI Image Generators
First, let’s consider what I’ve termed ‘AI Image Generators’. These are the tools that I think people are most concerned about. You give it some images that are similar to the image you want to create, you tell it what you want it to do, press ‘go’, and then you get an image.
The results can be impressive too. Take the example below by artist Jason Allen.
Now, you may think that it isn’t actually that impressive because an AI-generated it, but I know that I couldn’t create anything like that – even with my experience with Machine Learning. Certainly not without investing considerable time and effort.
The “just set it up and press ‘go’” description… When you put it like that, I can understand why people are nervous or sceptical, but in reality, it’s not that simple. I know from experience that it’s not a trivial process to get a Machine Learning application to do what you want. It takes a lot of training, tweaking, and trial & error.
For example, I was talking to a friend about a project he was working on. His girlfriend is a massive video game fan, and he wanted to make some ‘trading cards’ of various characters made to look like his girlfriend. So, he started collecting images of characters, as well as photos of his girlfriend, and training the AI to create the images. He ran into all sorts of problems. The characters’ hands and feet were weird shapes and had the wrong number of fingers and toes. Others had extra ears or too many arms. Many of the poses were unrealistic, with incorrect body proportions or contorted into impossible positions (see movie posters for any film with a female action hero!). It took him days to train the tool, and he had to discard thousands of images to even get close to the kind of results he wanted (and then many more hours to create the final results).
And that’s part of my point. These tools may get easier to use over time, but they still require a fair amount of skill and effort to get reasonable results. As with any tool, it takes time to familiarise yourself with it and work out how to make the best use of it. An amateur isn’t going to be able to come along and knock out professional-quality images in a matter of hours!
Authorship
However, I can understand why people are nervous. Generating an image using AI raises questions about who the author really is. There is also another important question: what if my image was used to train the AI? Doesn’t that mean I contributed to the final result, even if they look nothing alike?
There are lots of stories about people ‘scraping’ images from Getty Images or other sites to create their images. And I agree that sites should probably have ways of allowing artists to opt out of having their art used in such a way without their permission.
But where do you draw the line? Most writers start out by trying to ape the style of their favourite authors. Most artists start by tracing or copying their favourite works of art. If it’s OK for a piece of art to influence my work, why is it not OK for the same piece to ‘influence’ the AI? Perhaps this is too deep and philosophical a question to tackle right now!
I do agree that it is probably not fair to compare AI-generated images to traditional artwork created by hand by human artists, without some indication or declaration that an AI was used. However, just as AI can be trained to create these images, it should be able to detect them, too. This will allow sites to determine which artworks were created by AI and which were made by real people. I think sites like Artstation are right not to flat-out ban them, and I hope that such sites are able to adapt to this new technology.
May we ban them?
Not so fast. Even if you don’t like these tools being used to create a generate a piece art, my feeling is that such tools can still have their place.
Let’s say I wanted to generate an image of a character for a story I was writing. I could find photos of a couple of different actors/characters and mash them together using AI to create an image to help me better describe the character. What if a concept artist needs to generate lots of different spaceship concepts quickly? Such a tool could help them do that faster. If it’s for personal use, rapid prototyping, or you aren’t selling the end result, then where is the harm?
Finally, and perhaps most importantly, I want to reiterate that an experienced artist is always going to get better results than anything that an amateur could produce. Like any tool, people who master it will become more proficient and be able to achieve more impressive results. And if it takes skill and experience to create the best artwork, then I would argue that you are still an artist.
AI ‘Assistance’ Tools
These apps are, surely, less controversial and are the ones I find more interesting.
As a 3D Artist, I often have to find Textures – images that are applied to 3D models to make them look more realistic and detailed. They have to be the right size and colour, they have to tile properly… it can be an arduous process. There are now AI tools that can generate textures from a list of instructions. For example, I can imagine telling the AI: “give me a brick wall where the bricks are x * y, every third brick is black and the rest are red”. It would then give me a texture I could apply to a model. I’m still building the model myself, but I’ve saved myself the time it would have taken to find or manually create the perfect Texture.
Alternatively, as a photographer, I can imagine going to a wedding and taking 2,000+ photos. Editing all those photos takes *ages*. But, if you had tools that could automatically detect and delete blurry images, or automatically apply your custom filters to every image – automatically adjusting it based on the image properties, then that would save you *a lot* of time. You’re still creating the filters, editing and cropping the images; the AI tool just makes your job easier.
Blurring the Lines
As with most things, there are some tools that blur the line between the two categories. For example, I used NVIDIA’s Canvas to generate this image:
I drew the image on the left, using a tool similar to MS Paint. The AI then generated the image on the left, based on the first image, its training and a series of parameters. It’s easy to see how such a tool would be useful to concept artists, game developers and many other artists.
Personally, I can’t see any problem with such tools, as they aren’t creating the final image for you; they are just part of the pipeline you use to create your art. These tools are simply empowering artists to work more efficiently. I sincerely hope that such tools can be adopted in places like the Games Industry, where creating detailed, realistic 3D environments is becoming increasingly time-consuming, as doing so will relieve some of the huge amounts of pressure currently faced by artists in that industry.
Final Thoughts
On balance then, I think I am in favour of such tools. I think sites like Artstation are right not to flat-out ban them, though I hope their use is properly regulated.
In my mind, it is no different from when the loom was introduced to the world of weaving, the graphics tablet to the world of comic books, or the camera to the world of portraiture. There are reasons to be concerned, but the outcry – as always – far exceeds the actual threat posed by the technology. Yes, there will be problems and controversy, but any technology can (and will) be abused. Like Deep Fake technology, it needs to be properly discussed and regulated. But that doesn’t mean we should try to ban its use altogether.
Ultimately, these apps are here to stay. They are powerful tools that will enable artists to work more efficiently. Some artists will struggle to adapt to the new technology, but that’s OK. It won’t kill off artists any more than the loom, the graphics tablet or the camera. The art world will evolve, as it always has, and the technology will find its place in that new world.
I also believe that AI tools will enable great artists to create some truly amazing results, and I, for one, can’t wait to see them.