After signing up for several waitlists, and deciding I really don't want to download any more apps, I've finally gotten sucked into playing with one of these AI powered image generators you've been hearing so much about. Microsoft has partnered with DALL-E for the Image Creator, which has the benefit of being browser based and a lot less popular than Midjourney right now, so the wait times are shorter. I also have a lot of Microsoft rewards points, which I've never found anything interesting to cash in on, so they're going towards "boosts" to shorten the wait times even more.
And… the results are exactly as advertised. Hands look weird. Lines don't quite match up. However, as someone who has dabbled in visual arts, and knows the amount of work that can go into similar images, the results are often astounding. I didn't spend much time generating photorealistic images (pool floatie in the shape of a Honda Accord!), instead trying out a lot of fantasy art in different styles. There was very little that came out so perfectly that it looked like it was intentionally drawn by a human being, without all the little telltale signs of AI art. However, I was frequently startled by how perfectly it managed to imitate other artists' styles. The program blocks out the use of most recognizable intellectual property and any prompts that it deems harmful or malicious. However, there are characters that have simply become too universal to filter out, like Elsa from "Frozen." I used her as the common variable for a lot of prompts, generating pictures of Elsa in the style of John William Waterhouse, Jean Honore Fragonard, Jean "Moebius" Giraud, Alphonse Mucha, Kathe Kollwitz, Erté, and Frieda Kahlo. Some turned out much better than others. Ever wanted to know what Batman would look like if painted by the post-Impressionist Henri Rousseau? Now you can find out.
Prompt construction takes some effort to learn, and you have to experiment with word combinations for a while before you figure out what the program needs to generate the image you want. Words like "corpse" and "kissing" trigger the filters on violent/sexual images, though not consistently. You don't get a specific style or medium unless you ask for one. You don't get a background unless you ask for one. Unless you specify "full figure," the default will always be a portrait view of any person or character. Even under the best circumstances, the results are very hit-or-miss. Amusingly, using the word "cute" with "cartoon" or "illustration" is almost always going to get you a chibi anime kid. Asking for something in a specific artist's style is a good shortcut, but not foolproof. I asked for a picture of a fairy in the style of Maurice Sendak, for instance, and got back four lovely pictures that looked like they'd been plucked out of children's books - but none of them remotely in Maurice Sendak's style.
As a consumer of art, I can see how this could quickly become addictive. You can generate just about any image you could think up with minimal effort, skipping the laborious efforts of a human artist, and not having to worry about copyright issues. The implications, frankly, are scary to think about. As a creator of art, it can be absolutely maddening to look at some of the images, because you have such limited control over how they turn out. My immediate instinct with so many of these pieces is to fix the glaring errors I see. (In addition to weird hands, DALL-E has a real problem with swords.) I get no feeling of accomplishment when I generate one of these images, because I know I did none of the work involved, and know that I would have executed them very differently. However, I can see how this would be a helpful tool for artists, especially at the conceptual stage.
As these image generators keep getting better, and the proliferation of AI generated images becomes more widespread, it's clear there's no putting the genie back in the bottle. Artists are rightly concerned about being edged out, and having their work devalued. As for me, I'm fascinated by these programs being able to accomplish in seconds what it's taken me decades and decades to do so much more poorly. They may never be a substitute for a real artist with real talent, but they're certainly doing better than most of us.
---
No comments:
Post a Comment