I’ve been experimenting with artificial intelligence (AI)-powered art generators and have become fascinated with them. This is due not only to their ability to generate incredible art — including very realistic-looking photographs and stunning painting-like images — but also because of the possibilities and issues for business and society that could arise as they become mainstream.
AI art generators combine machine learning (ML) and natural language processing to generate images from natural language text prompts input by users. Their ML models are based on the Generative Pretrained Transformer (GPT) model initially developed by Open AI (in 2018), and which was trained on millions of images harvested from the Internet. Open AI has further developed the model to where it is now at version GPT-3.
One of the (many) really cool things about AI art generators is that they do not generate images by simply mashing up existing images. Instead, they use a diffusion modeling technique whereby a random number of pixels are manipulated over and over until eventually they are assembled to match the description of a text input, resulting in the generation of an entirely new image.
There are a number of AI-powered art generators. Some are still in beta or are available for use on a limited basis; some are free; others require a subscription. Some place restrictions on the kinds of images you can generate and how you can use them (more on this follows). Moreover, the technology is advancing quickly and new versions of products are appearing regularly.
Open AI’s DALL·E 2 currently gets most of the attention due to its ability to generate photo-realistic images. I’ve not yet had a chance to use DALL·E 2. (I signed up several months ago but haven’t been approved or whatever.) But I have used others, including LLC’s Craiyon, Hotpot.ai’s AI Art Maker, and Stability AI’s DreamStudio Lite.
Using an AI art generator is quite amazing. There’s really no other way to describe it. In fact, the ability to brainstorm or ideate with these tools is addicting: with each new image generated you come up with new ideas, which, in turn, lead to more new ideas and yet more new images. Think of it as a tool for automated iterative art generation.
Figure 1 shows an image I created using Stability AI’s DreamStudio Lite (beta). To generate this image, I entered the prompt “locomotive crossing the prairie on fire” into the tool. The idea for this image evolved from an earlier attempt, which resulted in a pretty boring photo of a train crossing a field (which was generated by me entering “locomotive crossing the prairie). I doubt that anyone would call the image in Figure 1 boring.
The image in Figure 1 appears here exactly as it was generated. I didn’t do any post processing (e.g., cropping, sharpening, noise reduction, tone correction). To make sure that this was, in fact, a novel image, I did a visual search using Bing image search to see if I could find it on the Internet. I couldn’t.
Quite simply, the results from DreamStudio Lite are very impressive. My image is photographically correct. The subject is not too centered (it’s placed according to the “Rule of Thirds”), and the horizon, sky, and foreground are unevenly spaced, which serves to add tension to the photo — all good aspects of a well-composed image.
Extensive Possibilities, Various Concerns
Using an AI art generator, it doesn’t take long before you start getting all kinds of ideas about how the technology might impact art, business, and society in general.
The Ability to Easily Generate Synthetic Media Will Lead to All Kinds of Opportunities and Amazing Art
Marketers, advertisers, graphic designers, and even novices and amateurs will be able to brainstorm and generate new image designs for all kinds of uses. But what impact will this have on professional graphic designers? I think that the effect will be measured. Yes, amateurs and novices will be able to create some amazing images. But AI art generators will also allow professionals designers and artists to take their skills and work to a whole new level. But this will require gaining experience with the tools.
I realized this soon after experimenting with AI art generators: if you want a tool to generate better, more impressive, images, you — the user — need to write more complex prompts (and many of the tools provide advice on how to do so). In addition, other user-defined criteria and tool functionality can affect the quality and style of generated images. For example, DreamStudio Lite gives the user the ability to select different model types for the tool to use and to adjust other functions, including how many steps the tool will spend diffusing (i.e., generating) your image. A configuration scale lets you adjust how much the generated image will match your prompt.
When using various AI art-generator tools, I noticed that for every good image I generated, I created about five to 10 unimpressive or sometimes even strangely weird or disturbing images. I believe that the cause of this was my poorly constructed prompts and some of the crazy, impractical ideas I fed the tool — largely due to my inexperience. In other words, AI art generators, like other design tools, have a learning curve.
The Ability to Easily Generate Synthetic Media Could Lead to Disinformation, Fraud, and Other Problems
What About Negative Impacts on Photographers?
The degree to which AI art generators will negatively impact professional photographers is difficult to predict. At first I was quite concerned. However, after considering the technology’s current limitations, I am less so. For one, the type, quality, and degree to which images created by AI art generators will match your prompts can be wildly chaotic. For instance, I was never able to generate the same image twice — even when trying to repeat previous results by reentering the same exact prompt.
A lot of photography involves people. Unless your subject is a celebrity, or at least well-known, it’s unlikely that an AI art generator will be able to generate a picture that includes your subject. Moreover, photographing people (well) requires capturing their emotions, often in response to their being in a specific location and/or in response to certain events (e.g., sporting) or situations (e.g., graduating or standing at the wedding altar). In other words, I don’t think that AI art generators do very well when it comes to real-world situations, at least in their current form. Consequently, I hardly think their rise signals the writing on the wall for photography.
I’m More Up in the Air About the Effect of AI Art Generators on Stock Photo Agencies
This is tricky. Photos of unknown people in various settings (e.g., playing in a park, couples on a park bench, riding bicycles, laying on a beach) are popular for advertising, websites, and other uses. But if you can use an AI art generator to auto-generate realistic images of such generic subjects in such generic settings, will you need to license them from a stock photo platform?
What About Copyright for AI-Generated Art?
When DALL·E first appeared, there were questions about who should own the copyright to images generated by the tool. This was logical, being that the models for such tools were trained on millions of images scrapped from the Internet. However, Open AI recently changed its policy around commercializing DALL·E output, now specifying:
I expect that other companies offering AI art-generator products will follow Open AI and give creators using their tools rights to commercialize the images they create. This makes sense. It’s just very difficult to see how AI art-generator providers could successfully license their tools to commercial enterprises if they wanted to retain copyrights to images created with their tools. Moreover, I always figured that AI art-generator providers would just follow the way of digital camera manufacturers who don’t make any claims to copyrights of images created with their products.
(For the record, the image in this Advisor was made using Stability AI’s DreamStudio Lite. Regarding copyright, the image falls under the Creative Commons 0 CC0 1.0 Universal Public Domain Dedication.)
These are some of my thoughts on my initial experience with AI art generators. No doubt about it, these tools offer exciting possibilities for novices and professional to create incredible art for many uses. I also see possible issues for business and society arising as their use becomes more mainstream. That said, I do not believe that they will prove to be the “be-all, end-all” art design tool. And certainly, I do not see AI art generators replacing professional designers and photographers.
But what is your opinion on the development and use of AI art generators? In particular, what impact do you see them having on the art world, commercial design, advertising, and other business uses? As always, your comments will be held in strict confidence. You can email me at email@example.com or call +1 510 356 7299 with your comments.