It has been several months since AI image generators were introduced and, as far as I know, no painters have lost their jobs, art is still not dead, and the current incarnations of AI have not yet caused any fundamental changes to how painters work. Of course, this is only the beginning. The “game changer for artists!!!” headlines may have subsided but AI continues to evolve at a breakneck speed. I've been monitoring the latest AI news and sifting through all those clickbait headlines so you don’t have to (you’re welcome) to find any interesting developments in regards to AI and painting.
You probably already know about the upgrades to Dall-E and Midjourney. While impressive, these changes are mostly refinements (like making it easier to iterate on a single image or to generate more accurate text.) A more significant development with AI in general is that ChatGPT can now see, hear and speak. Take a picture of your open refrigerator, for instance, and ChatGPT can list recipes based on the ingredients it sees. Take a picture of one of those overly-complicated parking signs that list a hopeless tangle of time restrictions and ChatGPT will tell you whether it’s safe to park there or not. You can also have a natural voice conversation with it now too. In fact, it was so natural that after using it for only a short while I found Siri to be almost unusable.
But for me, the most interesting developments in AI generation have been the experiments in the way we collaborate AI with image generators. Text-based image generation is clunky because it’s difficult to describe a scene to a human, much less a computer. And then there are those small but significant delays between each prompt and iteration— you write a prompt, wait for a set of images, modify the prompt, wait for more images, over and over until you get what you want. Or not (getting an image that is 80% there is easy but that last 20% is proving to be a big problem.) When making a traditional painting, feedback is immediate and constant. “Does this mark work? Nope. How about this one? Or this one ? Yes, that’s better.” Etc. Each painting contains thousands of micro-decisions, iterations and experiments that come directly from the artist and happen in real-time. With the AI prompt process, the delay and lack of control makes you feel more like a curator than a creator.
Which brings me to real-time AI painting where the AI paints along with you, interpreting each mark or change you make. Let me show you what I mean. This is a quick video demo I made using the developer's starting default prompt:
You can try it out HERE.
And here's a roundup of some interesting demos a came across recently:
Controlling the size and shape of watercolor washes:
Adding movement to still images
And here is a more sophisticated version of real-time AI painting:
Thanks for the post. Fascinating.