- The Tynan Files
- Posts
- Will the last artist left on Earth please turn out the lights?
Will the last artist left on Earth please turn out the lights?
AI image generators like Ideogram have gotten good enough to replace humans. They're also hilarious and terrifying.
Meet Sister Snoop. She's one bad mutha-superior. Source: Ideogram/otenrab.
AI tools like ChatGPT and Google Bard are very good at stringing words together in competent, grammatically correct fashion. But they're not writers. I can always tell when a chunk of text has been created using one of these tools, because it's boring and predictable. Their writing has no personality, because, well, they're not persons.
AI-image generators, on the other hand, are incredibly sophisticated. The work they generate in a matter of minutes is impressive and even a little frightening. And though they are far from perfect, tools like Midjourney and Stable Diffusion may prove to be an extinction-level event for illustrators, graphic designers, sign makers, poster artists, and other people who get paid to create commercial images.
The terrible irony here is that these very large AI models were created by feeding them many millions of images created by humans. And now, they're good enough to replace (most) of these people. Ain't technology grand? [1]
Now there's a new AI artist in town, birthed by a handful of exiles from the Google Brain project. What sets Ideogram.ai apart is its ability to incorporate typography inside images — something other AI image generators struggle with. And while it's far from flawless, it's right often enough to make it a viable alternative to hiring a human artist to do the same.
I hear ya pal. Save room for me on your next visit. Source: Ideogram/Mico11
Art for art's sake, money for god's sake
A lot of the early talk about Ideogram is that it's going to displace Midjourney as the AI image generator of choice. I'm not so sure about that. I spent a stupid amount of time last night generating images on both engines using identical prompts, and there are some significant (and occasionally hilarious) differences.
For example, here's an example of "Margot Robbie as Barbie zombie" on Ideogram:
Here's one of the variations produced by Midjourney [2]:
While I would happily let either of these hot zombies eat my brains, I think Midjourney's is more cinematic. Also, Ideogram's Margot is almost a dead ringer (sorry) for her Suicide Squad character, Harley Quinn.
This is what Midjourney does when I ask it to draw "Elon Musk as The Wizard of Oz":
And this is one of Ideogram's attempts:
That's more like The Wizard of Hoss [3]. Looks like Cowboy Elon needs to spend less time at the chuck wagon.
One of the tests I like to do with these things is to ask them to draw in the style of a certain artist. Here, for the most part, Ideogram did a better job than Midjourney. For example, here's one of Ideogram's efforts to create an image in the style of Edward Hopper.
And here's the best that Midjourney could do:
Now here's Midjourney trying its (virtual) hand at Pop Art icon Roy Lichtenstein:
This is Ideogram's best attempt at it:
In addition to being much closer to Lichtenstein's comic book style, Ideogram also managed to incorporate the text bubble I asked both of them to draw, though it mangled two of the words, and it has the wrong person saying it.
Fasten your seat belts folks, it's about to get weird
I then decided to throw the names of some underground comic artists at these engines. The results were mixed at best. I tried Basil Wolverton and Harvey Kurtzman (who created some of the iconic Mad Magazine covers of the 1950s) and Ralph Steadman (most famous for his work with Hunter S. Thompson).
It was, frankly, disappointing. This is the best Ideogram could do with Steadman:
Then I tried R. Crumb, the famous artist behind Zap Comics. Ideogram did a decent job of capturing his style. Here's one:
Apparently, Donald and Melania have adopted a young Stephen Colbert.
And here's another:
In which D. Trump has three heads, with Nurse Ratched looking on.
And then I asked Midjourney to create the same image. This is the exact prompt I used: "the trump family as drawn by R Crumb, illustration."
Here are the first four images it produced:
Looks like a group photo taken at the last CPAC convention.
Talk about your "AI hallucinations." Midjourney clearly has developed its own ideas about the 45th POTUS, and it's not holding back. [4]
My final prompt was this: "Donald Trump and Melania Trump portrait in the style of Basil Wolverton, photo, 3d render."
I am without words. I guess a picture really is worth a thousand of them.
Does this make you want to use tools like Midjourney and Ideogram more, or less? Register your opinions in the comments below.
[1] The standard response to this from the AI crowd is that "this technology frees artists (writers, programmers, fill in the blank) from mundane/repetitive/low-level tasks, allowing them to take on more innovative, exciting, strategic projects, yadda yadda." Yeah, maybe for the top 5 or 10 percent. Everybody else will be wearing hairnets and flipping burgers. At least, until they're replaced by Flippy, the Burger Flipping Robot.
[2] Both Ideogram and Midjourney produce four images at a time, some dramatically different from the others. I'm subjectively picking my favorites here.
[3] That's a "Bonanza" reference, for anyone born after 1970. Sing it with me: "We got a right to pick a little fight, Bonanza! If anyone fights anyone of us he's gotta fight with me!"
[4] It's not wrong.
Reply