why no AI

I thought I would take some time to write the reasons I think it is worthwhile to consider a human artist over generative AI art.

I’ve worked on a number of projects where AI art is woven-in, or replacing human art. I believe AI art can lead to problems that I don’t think were anticipated.

My intention here isn’t to explore the wider debate surrounding AI. This page is just about what I know, and my experience, within the world of producing artwork.

Obviously I have a horse in this race - I am an artist, and I directly compete with AI artwork. This is entirely biased, but, if you’re on the fence, this is the perspective of one artist ~

• AI deliverables are difficult to work with. They are typically flattened, rasterised, and compressed. This is often good for final export, but it can become convoluted, and sometimes expensive, to retroactively turn AI produced files into workable source files. Humans naturally foster these files as they work.

• A collection of AI images is a mood board. You may want to pass these files along to animators, riggers, web designers, games designers. Without the source files, it’s likely they can’t be worked with unless reconstructed, which is exactly the cost that was being avoided.

• Small changes via prompts introduce inconsistencies and new artifacts. AI is trained largely on images with compression artifacts, and so will often produce images with those artifacts. Humans can make small changes quickly by using their source files, and don’t need to reproduce the entire image.

• AI continues to struggle with vector artwork. While AI is proficient at generating raster artwork and video, it remains poor at producing vector artwork. Vector art is crucial for creating scalable and adaptable artwork, particularly where artworks are needed for branding or user interfaces.

• AI often confuses and blurs styles. For instance, despite being prompted to be ‘flat’, AI is known to introduce shading and texture to objects it has mostly been trained on appearing shaded. I’ve seen AI produce sets of icons that subtly change style between pieces. It isn’t always noticeable at first glance, but it matters where consistency is important.

• AI is commonly used for conceptual work, but I believe this is backward. Conceptual stages are arguably when you want the most creative, unusual, and human designs to carry your project forward. Polished artwork at early stages is making thousands of small decisions for you, which muddy the intention. Scrappy sketches are much easier to work with - they reveal exactly where the mind of the project is at, and trim away everything except what’s important.

• AI art steers towards the average. AI is trained, overwhelmingly, on the most common version of things. It works, in large part, by predicting a most likely outcome. Often we want the most likely thing (the most likely response to ‘3 teaspoons of baking powder in grams’ for example). Art that is predictable is ignored. Art should always be (at least a little bit) surprising. Art should express, as well as achieve, its intention.

• An AI’s audience is wide, and it produces artwork to match. Designers and illustrators are fans of art, they have studied it and surrounded themselves with it for most of their lives. That also describes a lot of people who don’t necessarily work in the field. It may not describe you, but it might describe a greater portion of your audience than you realize.

• AI art is obvious. It remains one of the most easily identifiable forms of generative AI, and least tolerated. ‘Obvious’ once meant six fingers and smeary text, but nowadays tells are stylistic choices, cliches, and overdetailing - things that can’t be solved with more tokens. There will always be someone who is able to tell.

• How we use AI reveals what’s important to us. AI imagery is an aesthetic that flags a project that would cut corners where art is necessary, and views art as functional rather than expressive or interesting.

• But, human artwork doesn’t need to be expensive. Simple, expressive pieces remains cost-effective, which happens to be the kind of work where AI struggles. I love awkward, imperfect artwork - human imperfections reflect where an artist has trained, what they enjoy drawing, how they spend their time. I’ve certainly produced my fair share.

• I’d spend some of the money you give me on sandwiches. We have a bakery up the road that does a great spicy salami and cream cheese sandwich, which I am interested in investing in further.

That’s my two cents.

I’m aware that some power users have access to tools and resources that may overcome a number of the above, but - maybe unsurprisingly - that doesn’t yet describe anyone who opts to work with me. Most people use the widely available tools.

AI moves quickly, and I’ll update this page as my understanding changes. If you think I’m wrong anywhere in the above, feel free to reach out, I’d love to hear from you.