The tech world never lacks trending topics, but one of the most interesting this year is the use of artificial intelligence (AI) to create art. The resulting images can be everything from grotesque to stunning.
Here’s an ultra-simplified explanation of how they work: Take millions, if not billions, of captioned images, and generate something new and unique based on a text description you provide, called a “prompt.” (This article(Opens in a new window) provides a more detailed breakdown of the process.)
Recently, AI tools have created memes (read about Loab, the “AI art cryptid,”(Opens in a new window) for some genuine chills). They’ve generated the imagery of an entire sci-fi short film and a video game and even won art contests.
Some call AI art a whole new artistic medium(Opens in a new window). Arguably the most popular is Dall-E, which is now used to create as many as 2 million images per day(Opens in a new window) alone. Few safeguards exist against using these AIs for nefarious purposes (think propaganda and disinformation). But that’s not going to stop people, especially with the truly open-source options.
“Pitbull reading on a laptop”
(Prompt by Eric Griffith; Generated by Dall-E)
Several big-name tools that were in private beta have become available to everyone, including the aforementioned Dall-E as well as Midjourney and DreamStudio. New mobile apps such as Wonder, Dream by Wombo, and Starryai are also available. Even big tech companies—namely Meta and Google—are in on it. Both have announced tools that will go beyond still imagery and make AI-generated videos, named Make-A-Video and Imagen Video, respectively.
Google’s Imagen will also eventually generate still images, but neither it nor Meta’s text-to-image tool is publicly available yet. Both companies know they’ll face an avalanche of criticism for the biases inherent in how the images are generated(Opens in a new window).
As the big companies hesitate, it gives smaller, privately owned AI art generators a chance to show off their wares well ahead of the competition. Some have heavy parental-type restrictions and content policies in place to prevent issues; some can be circumvented(Opens in a new window). At least one that we tested (Wonder) popped up some unexpected, full-frontal female imagery—tastefully rendered, but still not NSFW. It’s also a hot-button issue for artists(Opens in a new window), some of whom are getting copied by the AIs(Opens in a new window). They may be losing their livelihoods.
Some of the imagery from these AI tools is breathtaking. There’s a reason that a person won a state fair art contest using Midjourney. It’s still a lot of work—he spent 80 hours honing his art prompt, plus he still had to use extra software tools like Adobe Photoshop and Gigapixel AI(Opens in a new window) to enhance the original AI image.
After seeing some cool images generated from next to nothing, we tested the top five AI art generators with free access versions to see what they could generate using the same prompts: Dall-E, Midjourney, DreamStudio Lite, Craiyon, and Wonder. The result is a direct, if subjective, comparison. Read on to see the stunning effects.
How We Tested AI Art Creation
We put out a call to PCMag staff for art prompts. Six people responded gleefully, sending in some nonsensical things sure to challenge any artist, living or digital. The goal was the use exactly the same verbiage on each AI, with no changes and little or no filtering. Thus, we didn’t add anything to the prompts to change the outcome (such as putting “photorealistic” or “8K” in a prompt, or clicking any filter/style options, for example).
The results are below in six slideshows, coupled with their queries and our subjectively picked winners. All images are the first results we saw for each prompt on each service. We generated no variations, which would have undoubtedly improved many of the images generated. The overall winner is listed below, with more details about each AI art generator.
1. “A pug driving a Mini Cooper underwater”
—Editor-In-Chief Wendy Sheehan Donnell
For Wendy, the choice was Midjourney. It was, after all, the only one that nailed the “driving” aspect of the prompt. Not that the others didn’t have lovely little pugs galore. But Midjourney’s default leaning into the artistic rather than realistic right out of the gate may have helped.
2. “Landscape with melting laptops and skulls in the style of Salvador Dali’s ‘The Persistence of Memory‘”
—PC Labs Director and Executive Editor John Burek
The former Dall-E mini wins this round. John says, “Craiyon definitely got the closest match/most literal interpretation to the original vision, for sure.”
3. “Bjork dressed as a goth playing golf, about to tee off at the ninth hole on a sunny day in Iceland”
—Deputy Managing Editor Jill Duffy
Midjourney wins again. Barely. But Jill is not impressed with any of the AI output. “Oh these are depressingly bad,” she says, adding that the first Midjourney image was okay, but it resembled a cemetery more than a golf course. Wonder is the only other AI that seems to really know what Bjork looks like, yet it still manages to render a face no one would want to see. (Some AI art generators, like Dall-E, prohibit the use of celebrity or politician images.)
4. “A rotund man with brown hair, a graying beard, and glasses using a Leica M3 camera to take a photo of the Chrysler Building”
—Lead Analyst Jim Fisher
Dall-E wins this one. Jim really wanted a realistic Leica but didn’t get one. He says, “Of the pictures Dall-E created, the fourth was my favorite. The art deco spire really evokes the spirit of the Chrysler Building, and while the photographer is most certainly not holding a Leica rangefinder, the camera looks like one that could be real.”
My question is: Why do almost all the AIs think a rotund photographer would wear a brown shirt?
5. “Tall skinny man playing guitar in a dead forest populated by owls and bats”
—Consumer Electronics & Mobile Managing Editor Eric Zeman
Dall-E wins. Eric says it was the “closest to what was in my head.” Too many of the images appear to him like they were either Jack Skellington or something out of The Walking Dead.
6. “Two scarecrows dancing the tango in a leafy field”
—Assistant Social Media Editor Jacqueline Goldblatt
Midjourney wins this one by going artistic again. Jackie and I both like the realism of some of the other images (except Wonder’s, which looks like a 3D render of a kid’s drawing, and not one of the good ones). But in this instance, with this prompt, artistic flourish definitely outshines photo realism. We also wonder why Midjourney seems to need therapy, generating some literal and figurative dark stuff for almost every prompt.
Overall Winner: Midjourney
So by the vote above, Midjourney wins as the best AI artiste today, but not in a landslide. Dall-E is a close second. But it’s all in the eye, or at least on the screen, of the beholder.
Anyone with AI art aspirations should try both; it turns out they’re the most discussed art generators right now for a reason. Between all the options above and a few that we didn’t even try, apparently there’s absolutely nothing you can’t make if you have the patience—and the right text prompt.
The AI Art Generators and How They Work
All the AI artistry is made approximately the same way—feed it words, and it spits out an image (or images). The tools can be subtly or even radically different in interface, pricing, and output. Here’s a quick breakdown for each of the five we tested above.
Recommended by Our Editors
OpenAI(Opens in a new window)‘s Dall-E is the darling of the moment. Named after both the artist Dali and the trash-compacting robot from the Pixar film WALL-E, OpenAI recently opened up Dall-E to everyone after months of waitlisting. Dall-E also does more than just take text prompts—you can also upload an image to get the tool’s optional edits.
You get 50 credits to generate a set of four images each when you sign up, and then 15 more credits every month for free. You can pay for more credits if you really get addicted. The max resolution for an image download is 1,024 by 1,024 pixels, but the edit function lets you add a “generation frame” to increase the overall image size.
Dall-E will soon be integrated with some Microsoft products such as the Bing search engine, where it will power an “Image Creator” feature.
You use Midjourney(Opens in a new window) within the confines of the uber-messaging-board Discord instead of on a stand-alone website. Turns out that is a great way to see what other people are creating via AI, in real-time. A mobile-friendly web-based app(Opens in a new window) lets you see your previously generated art as well.
The Midjourney interface in Discord can be dark and forboding, much like the art it seems to generate.
To make an image(Opens in a new window), go to one of the channels labeled “newbies” and type “/imagine” followed by your text prompt. You’ll get a 2×2 grid of 4 images that you can upscale resolution (the max is 1,664 by 1,664 pixels) or create variations of. You get 25 free “jobs” when you sign up, for free. After that, you pay.
The web-app DreamStudio(Opens in a new window) Lite is based on stability.ai’s Stable Diffusion(Opens in a new window) (technical types might want to try it locally on a PC that can support it). DreamStudio is the way most of us will use it. It’s also accessible on a mobile browser. Try DreamStudio for free with about 200 generations; after that, it’s approximately a penny per AI pic.
The interface looks deceptively simple, but the sliders provide a lot of amount of control. There’s an excellent prompt guide listing all the tricks and tips you need to write a good one. The engine by default only shows one image at a time. You can change that setting to generate up to nine at once. The base image is 512 by 512 pixels, which can be doubled when you want to finalize it.
Craiyon was originally called Dall-E mini, for it was an attempt to do exactly what OpenAI does with an open-source model. It’s completely free to use (it displays ads). You don’t have to sign up.
It also has the simplest interface, just a text box on the Craiyon.com(Opens in a new window) page, so it works on mobile devices (there’s also an official Android app(Opens in a new window)). Click the crayon icon after you enter the prompt and in less than two minutes (usually) you get nine samples. You get to watch a commercial while you wait! Downloaded images are small, only 256 by 256 pixels, with no option to upscale or make a variation.
This AI art generator from Codeway Dijital(Opens in a new window) comes in app form only, for both iOS(Opens in a new window) and Android(Opens in a new window). You get 3 days to try it for free, then you start a paid weekly subscription. Heavy users will want the $ 29.99-lifetime purchase.
With Wonder, it’s not enough to put in a text prompt; you typically also pick an AI Model (a style). You can select “No style” if you want a surprise. On each image, there’s a re-create button to make a variation. If you want a more distinct variation, go back and pick a new AI model. Give images a thumbs up or download them if desired. Items not saved, shared, or “published” on a paid account are lost forever. All pics come in portrait orientation at 1,024 by 1,536 pixels.
AI Art Is Fun…for a While
In the end, the art created by the above AIs is still more a conversation piece than anything. It’s fun to play with a prompt and try to get it to deliver something approximating what you want. If you can get it to show you exactly what you pictured, the AI might be training you.
The fun dissipates when you have to start paying. If you’re using it for actual work, you would be better off working with a living artist with whom you can have a mutually beneficial relationship.
Plus, consistency isn’t a strong point for the AIs. It’s not like you can create a series of works about, say, the same character or even locales, enough to match them up to tell a sequential story. Comic book artists seem to have little to fear, though the photo-realistic backgrounds might be killer.
The better a person gets with a prompt written just the right way, maybe those inconsistencies will fade. But it’ll never feel as satisfying as creating something yourself.
Get Our Best Stories!
Sign up for What’s New Now to get our top stories delivered to your inbox every morning.