Now in its 16th season, RuPaul’s Drag Race has birthed more than a few iconic lip-sync battles, but precious few have featured Muppets. None, even. AI Drag Race changed that. In the Instagram account’s recent season finale, Miss Piggy, wearing an AI-generated drag look, faced off against lover-turned-rival Kermisha Ihman, who had a thick, 40-inch-long ponytail atop her green felt head.
Tackling Lady Gaga’s “Telephone,” the two whirled and jumped, kicking and bucking in front of head judge Betty Boop. Kermisha worked her faux-nailed webbed feet, sickening in her bejeweled purple corset, but ultimately she fell to Piggy, whose fringe flew as she went for a well-timed jump split at the song’s climax. As the commenters watching the video noted, “Kermisha fought hard and devoured,” but “Miss Piggy ate.”
AI Drag Race is just one of a burgeoning number of AI-generated accounts popping up on Instagram and TikTok, with creators not only pitting their favorite fictional characters against each other in seasons meant to mimic the original show, but also creating and generating their own queens.
There’s AI Horror Drag Race, which features characters like Pennywise and Billy the Puppet facing off against Ghostface from Scream. There’s also Big Girl’s Drag Race, which is exclusively for curvy characters. (Past winners include Ursula from The Little Mermaid and Fat Albert.) All the participants in Fantasy Drag Race’s seasons look vaguely like dolls, while Slay Drag Race AI, which recently eliminated Scooby-Doo sidekick Shaggy and Marceline from Adventure Time, is hosted by a surprisingly accurate-looking Dora the Explorer.
Some accounts put their participants through signature Drag Race challenges, like Snatch Game and commercial shoots; others create not just the contestants’ runway looks but also their street-clothed everyday looks for workroom shots or interview stills.
They’re also operating in something of a copyright gray area. Drag Race, Miss Piggy, and Pennywise all have stakeholders. Creators behind the various accounts claim that what they’re doing is a parody, or falls under the fair use doctrine of the Copyright Act, which stipulates that “transformative” works are protected. Beyond that, though, are questions about whether AI should even be able to generate an image of, for example, Homer Simpson in a kitsched-up showgirl costume. One day, a court may decide that AI models can’t train on copyrighted characters or produce outputs depicting them, says James Grimmelmann, a professor of digital and information law at Cornell University.
“I think that would be a loss to culture, but it’s also something that could be consistent with existing copyright law,” Grimmelmann adds. “We’re in a period of legal uncertainty right now.”
Michael, a fashion illustrator who lives in Europe and runs The Official AI Drag Race account, says he first came to spoof Drag Race a few years back after finding Paper Drag, an online competition in which participants are tasked with drawing a look based on a theme each week. A fan of extravagant costumes, he’d already been drawing drag artwork in his spare time, so it seemed like a natural fit. A few years later, while walking his dog with his partner, the two started coming up with fake drag names.
“Then,” Michael says, “we found this simulator online where you can create custom queens and put in stats for them, like acting, comedy, dance, design, improv, runway, and lip sync, and give them a score from one to 15. We thought it would be funny to come up with seven queens each and run a fake season just to see whose queen won.” They did that about 10 times over the course of a year, taking personal pride over who won each season. (Michael and the other creators interviewed for this story declined to give their full names, and some chose aliases, citing fears of harassment or copyright takedowns.)
Around 2022, Michael says he started to see more AI images online and began to dabble with Craiyon, creating images of queens. He and his partner decided to take one of the seasons they’d already simulated and create AI images to accompany it, using the existing simulation to decide who’d be knocked out of the competition. He put it online and told some of his friends. Now his account is one of the most popular AI drag Instas, with 13,000 followers and counting.
To craft his looks, Michael first comes up with a character like Tina Donna, a competitor in his current season who hails from Nashville. “She’s the local girl who managed to get on TV,” he explains. “She’s from a small Southern town originally and she doesn’t have as much money as everyone else, but she has a winning personality.”
Pulling from past real-life Drag Race episodes, he determines a runway theme and then plugs a description into Copilot of what he thinks each particular queen would wear, writing something like “full-body image of a plus size [Latinx] drag queen with flowing blonde hair and an ornate star-covered headpiece posing elegantly on RuPaul’s Drag Race main stage,” and then adding details about her dress, any accessories he wants, and what kinds of shoes he sees her wearing. He tweaks each image that’s produced, adding in new prompts to get it to where he wants to be and ideally making the queen’s face look at least a little consistent with how it has looked in weeks past.
“Sometimes I get the render I want the first time, but sometimes it takes me 50 times of running the same input over and over again until it gives me the magic output,” Michael says. He uses his fashion expertise to produce costumes that he thinks are relatively realistic, using descriptors that help hone the looks’ silhouettes and textiles, as well as the techniques theoretically used to create them. “We like it when someone is scrolling Instagram and thinks, ‘Wait, what season of Drag Race is that?’” he explains.
There are still limitations to what AI can do. Many creators note that inputting “drag queen” into Copilot almost always yields a showgirl-like skinny white figure wearing a high-low dress, legs exposed in the front, with a big train. Getting a plus-size queen or a queen of color can require some finessing, as can getting any sort of specific makeup look. (This simplistic and whitewashed idea of drag performers echoes how a lot of AI tools generate queer people.)
Boppy, the creator of Haus of Dreg, says they’ve had trouble getting a consistent Daphne from Scooby-Doo, noting they had to ask for “Daphne Blake from Mystery Incorporated” lest they occasionally end up with the cartoon dog instead. The creator of another page, SpongePaul’s Drag Race, has had a hell of a time persuading AI to put a wig on Big Bird.
When creators make multiple images of the same performers—for a spliced-together lip sync clip, let’s say—the limits (and reach) of AI can be unintentionally hilarious. Asking Copilot (Microsoft’s AI tool seems to be a favorite) to put a queen in a number of different positions can cause it to put less effort into the hair and garment, meaning you get wigs and dresses that get bigger and bigger and bigger, or that vary from frame to frame.
When it comes to eliminations, every account has their own way of doing them. Michael relies on the same simulator he has used since he was just messing around with his partner for the Official AI Drag Race, keeping his competition somewhat rigid and results occasionally shocking. Haus of Dreg’s Boopy has six of his friends judging. Horror Drag Race’s Shayne picks his winner and loser based on the likes and interactions their looks get. “You want it to seem fair so people can feel like they have a role in the production of the show,” Shayne says. “I don’t want it to seem manufactured, fake, or expected.”
On the other hand, part of what makes an AI-generated drag competition work is the show’s existing formula. After 15 years on the air, hundreds of episodes, and countless international spin-offs, RuPaul’s Drag Race has developed a familiar series of beats, characters, and story points. That’s part of the reason why AI creators know that, when they’re putting together their initial lineups, they need to have comedy queens, body queens, fashion queens, BIPOC queens, and queens of size. They can also often tell who’s going to be in it for the long haul, just based on character tendencies and how they would presumably perform in each challenge, or how well they’d fare under a judge’s scrupulous eye.
“What a lot of people like about the actual show Drag Race is its high-stakes competition factor,” says Más, who created Fantasy Drag Race. “Eventually, you learn to expect certain queens to never fall to the bottom or to expect other queens to always be in the bottom.”
It’s that set of expectations and tropes that makes AI drag work, since fans—and potentially the AI—know not only the look of the show, but also the language it uses. “AI drag is an accumulation of all previously done concepts and ideas,” Más explains. “That’s what makes it interesting, but also grounded in a way. The image generators are trained on videos and photos and web searches that contain the full history of drag, fashion, and pop culture, making what they create an accumulation of queer culture overall.” This may be a bit hyperbolic. Considering its history, it’s almost impossible to understand the full history of drag, but Más has a point: AI has a unique ability to synthesize ideas.
The blending of the real and unreal could explain why some followers of faux Drag Race get so passionate about what they’re seeing on their feeds. Michael says he “lives for the overreactions of fans” who believe his creations are real people. He says people often ask him for a queen’s actual Instagram handle.
“I also get the occasional hate comment from someone saying I’m taking away jobs from real drag queens,” he says. As an illustrator himself, Michael says he’s aware “that AI is coming for my job,” but doesn’t believe his Instagram passion project is taking money away from humans. “If someone isn’t going to the club and tipping a real drag queen because they saw AI Drag Race, that’s a problem with the person and not my Drag Race,” he says.
Fantasy Drag Race’s Más says she has gotten into scrapes with other creators in group chats, too, after questioning how seriously they were taking the whole process. “I’m a queer, nonbinary Mexican in upstate New York,” she explains. “Someone saying that my drag competition isn’t their cup of tea or that some look I made is ugly isn’t going to affect me at all.” Still, she says, it’s understandable that people get emotionally attached to her work.
Unfortunately, that kind of attachment also comes with a sense of looming dread, since the whole idea of AI-generated Drag Race is a play on a big franchise. While some creators argue that what they’re doing is parody, posting what Grimmelmann says are “almost completely useless” (or perhaps pointless) copyright disclaimers absolving themselves on their main Instagram page, others acknowledge that they’re likely building their followings on shaky ground.
A number of accounts, including one that featured exclusively Disney characters, have already been pulled off Instagram, giving creators who use only animated or existing characters more than a bit of pause. “I’m very scared of getting taken down,” says Haus of Dreg’s Boopy. “But if I did, then so be it. I mean, what could I even do?”
“I make sure that I don’t do anything to sexualize the characters, and I don’t do anything to diminish their actual tone,” Horror Drag Race’s Shayne adds. “I’m just merging two mediums—horror and Drag Race—and blending it up into something that both groups of fans can enjoy.”
It’s not just Drag Race fans that are enjoying the AI experience, either. Mhi’ya Iman Le’Paige, a queen from season 16 of Drag Race, just wore a look down the runway that first originated in a run of AI-generated images. One of her season 16 sisters, Plane Jane, follows at least one of the AI creators.
The Official AI Drag Race’s Michael says he has had multiple queens reach out asking to use their fictional creations as inspiration, with an unnamed queen from an international franchise asking Michael to design their entire package of runway looks based just on his Carla Montecarlo images. “I feel like it’s only a matter of time,” Michael says, “before I’m watching TV and spot something that I rendered a year ago.”
