AI ‘Dream Girls’ are Coming for Porn Stars’ Jobs

The Washington Post

14:38 JST, March 1, 2024

LAS VEGAS – Dozens of porn stars in lingerie lined the convention hall, but Tom Maupin, 72, knew exactly who he was there to see. The retired schoolteacher had flown from Omaha to Las Vegas, purchased an $85 ticket and waited in line with hundreds of other porn fans in hopes of getting some face time with actress Jill Kassidy, 27.

They’d been chatting online for months, but it can be tough these days to tell whether the sexy talk you’re paying for comes from the performer herself or an AI chatbot, Maupin said on the show floor of AVN, the adult entertainment industry’s annual conference. He was looking forward to their first in-person meeting.

Maupin approached the table where Kassidy, who uses a stage name, was signing autographs, and the two exchanged a hug. They talked for 10 minutes before Kassidy turned to her next fan, who gifted her a vape pen, and Maupin wandered away. Did she remember Maupin from their online chats?

“It’s hard to keep people straight,” Kassidy replied, smiling. She took a hit of the vape.

For Maupin, the real humans behind porn are the entire draw, but some porn companies are willing to bet that he’s a dying breed. Since the first AVN “expo,” in 1998, adult entertainment has been overtaken by two business models: Pornhub, a free site supported by ads, and OnlyFans, a subscription platform where individual actors control their businesses and their fate. Now, a new shift is on the horizon: artificial intelligence models that spin up photorealistic images and videos that put viewers in the director’s chair, letting them create whatever porn they like.

Some site owners think it’s a privilege people will pay for, and they are racing to build custom AI models that – unlike the sanitized content on OpenAI’s video engine Sora – draw on a vast repository of porn images and videos.

This vision for the industry’s future raises a host of difficult questions: How do you compensate performers whose likenesses are used to create AI content? Will consumers like Maupin be excited by AI porn at all?

But the trickiest question may be how to prevent abuse. AI generators have technological boundaries, but they don’t have morals, and it’s relatively easy for users to trick them into creating content that depicts violence, rape, sex with children or a celebrity – or even a crush from work who never consented to appear. In some cases, the engines themselves are trained on porn images whose subjects didn’t explicitly agree to the new use. Currently, no federal laws protect the victims of nonconsensual deepfakes.

“AI can’t replace [adult] performers, and people who say that are misunderstanding what performers do,” said Heather Knox, director of operations at Elevated X, a software company that helps adult performers manage their online brands. “But this technology will catch on, and it will get abusive before it gets helpful.”

Bespoke porn on demand

Adult entertainment is a giant industry accounting for a substantial chunk of all internet traffic: Major porn sites get more monthly visitors and page views than Amazon, Netflix, TikTok or Zoom, according to an academic analysis published last year in the Journal of Sex Research. The industry is a habitual early adopter of new technology, including VHS, DVD and dot com. In the mid-2000s, porn companies set up massive sites where users could upload and watch free videos, and ad sales footed the bills. As the 2010s progressed, adult stars mastered social media marketing, and many migrated to subscription streaming platforms rather than chasing contracts with major studios.

Now AI is here, and players across the industry are rushing to figure out what it means for business.

At last year’s AVN conference, Steven Jones said his peers looked at him “like he was crazy” when he talked about AI opportunities: “Nobody was interested.” This year, Jones said, he’s been “the belle of the ball.”

Jones is the owner and operator of a once-popular collection of porn websites. He entered the business in 1999 with a site dedicated to sexy snapshots of college students. He renamed himself “Lightspeed,” and, within seven years, the married father of two was operating a collection of more than 30 porn sites and making half a million dollars a month.

By 2013, it all fell apart. The self-described science-fiction nerd watched his revenue dry up after Pornhub – the tube site Jones refers to as the “evil empire” – started hosting free content. He ended up leaving the adult industry for a mainstream tech job.

Then the tides of fate turned again for Jones. His love for technology and shows like “Star Trek” made him an early believer in the power of artificial intelligence to change the world for good. Watching the public react to the releases of OpenAI’s conversational bot, ChatGPT, and image engine, DALL-E, in recent years, Jones felt an excitement he hadn’t known since getting his first computer as a teen, he said.

He called up his old business partner, and the two immediately spent about $550,000 securing the web domains for porn dot ai, deepfake dot com and deepfakes dot com, Jones said. “Lightspeed” was back.

But Jones’s plan to create consumer-friendly AI porn engines faced significant obstacles. The companies behind major image-generation models used technical boundaries to block “not safe for work” content and, without racy images to learn from, the models weren’t good at re-creating nude bodies or scenes.

One major model, Stable Diffusion, shares its code publicly, and some technologists have figured out how to edit the code to allow for sexual images. Particularly dedicated fans will come together on forums such as Reddit to share their nude creations, and some stand-alone deepfake creators make money churning out abusive nude images of people who didn’t consent to appear. But for non-techy laypeople wanting to experiment with a porn generator, the options were few.

So with help from an angel investor he will not name, Jones hired five employees and a handful of offshore contractors and started building an image engine trained on bundles of freely available pornographic images, as well as thousands of nude photos from Jones’s own collection. Users create what Jones calls a “dream girl,” prompting the AI with descriptions of the character’s appearance, pose and setting. The nudes don’t portray real people, he said. Rather, the goal is to re-create a fantasy from the user’s imagination.

The AI-generated images got better, their computerized sheen growing steadily less noticeable. Jones grew his user base to 500,000 people, many of whom pay to generate more images than the five per day allotted to free accounts, he said. The site’s “power users” generate AI porn for 10 hours a day, he said.

Jones described the site as an “artists’ community” where people can explore their sexualities and fantasies in a safe space. Unlike some corners of the traditional adult industry, no performers are being pressured, underpaid or placed in harm’s way, he said. And, critically, consumers don’t have to wait for their favorite OnlyFans performer to come online or trawl through Pornhub to find the content they like.

Next comes AI-generated video – “porn’s holy grail,” Jones said. Eventually, he sees the technology becoming interactive, with users giving instructions to lifelike automated “performers.” Within two years, he said, there will be “fully AI cam girls,” a reference to creators who make solo sex content.

Fighting abuse

It’s too early to call Jones’s venture a success. Right now he’s breaking even, he said, and all revenue goes toward improving the AI. As his customer base grows, so will the massive computational cost of generating images. It costs $12 per day to rent a server from Amazon Web Services, he said, and generating a single picture requires users to have access to a corresponding server. His users have so far generated more than 1.6 million images.

Meanwhile, lawsuits threaten access to training data, those troves of pornographic images bundled online. Copyright holders, including newspapers, photographers and artists, have filed a slew of lawsuits against AI companies, claiming that the companies trained their models on copyrighted content. If plaintiffs win, it could cut off the free-for-all that benefits entrepreneurs such as Jones.

What keeps Jones up at night is people trying to use his company’s tools to generate images of abuse, he said. The models have some technological guardrails that make it difficult for users to render children, celebrities or acts of violence. But people are constantly looking for workarounds.

Sometimes they’re successful. The prompt “adult woman with no breasts,” for example, may render a character that looks childlike. Jones said his team takes down images that other users flag as abusive. Their list of blocked prompts contains 1,000 terms, including “high school.”

“I see certain things people type in, and I just hope to God they’re trying to test the model, like we are. I hope they don’t actually want to see the things they’re typing in.”

Peter Acworth, the owner of kink dot com, is trying to teach an AI porn generator to understand even subtler concepts, such as the difference between torture and consensual sexual bondage. For decades, Acworth has pushed for spaces – in the real world and online – for consenting adults to explore nonconventional sexual interests. In 2006, he bought the San Francisco Armory, a castle-like building in the city’s Mission neighborhood, and turned it into a studio where his company filmed fetish porn until shuttering it in 2017.

Now, Acworth is working with engineers to train an image-generation model on pictures of BDSM, an acronym for bondage and discipline, dominance and submission, sadism and masochism.

“We’re not going to let [users] type in a sentence. It’ll be a series of checkboxes, like ‘These are the things I want, and this is the type of person I want,’” Acworth said. If his company makes the engine available to customers, he said, it would require “ad infinitum” testing and human moderators watching the results – both pricey and potentially unviable solutions.

Before OnlyFans was founded in 2016, Acworth said his company’s annual revenue peaked at $30 million. Today, it’s around $12 million and stagnant. From his vantage point, the entire adult industry is rushing to adopt AI – and he doesn’t want to get left behind again.

“We don’t want to be in a position where everyone else is able to spit out amazing content and we can’t.”

Anxiety among sex workers

For people in adult entertainment, arguing for your industry’s right to exist can be part of the gig. Both Jones and Acworth said AI porn is a desirable new option. But on the show floor of AVN, some dismissed the idea as a bad one.

“That’s just weird. It’s not a real person,” said Ben Schwartz, 34, who was standing next to a line of fans waiting to meet famous performers from the website Brazzers.

Others alluded to a porn apocalypse, with AI wiping out existing models of adult entertainment.

“Look around,” said Christian Burke, head of engineering at the adult-industry payment app Melon, gesturing at performers huddled, laughing and hugging across the show floor. “This could look entirely different in a few years.”

Performers themselves tend to take a measured view. If AI creates new ways to make money – such as lucrative licensing deals or personified chatbots that solicit tips from fans – sex workers will get on board, said Stella Skye, 32, after an AI marketing seminar sponsored by AVN. (Skye, like other performers in this story, uses a stage name for privacy.)

But the age of AI brings few guarantees for the people, largely women, who appear in porn. Many have signed broad contracts granting companies the rights to reproduce their likenesses in any medium for the rest of time, said Lawrence Walters, a First Amendment attorney who represents adult performers as well as major companies Pornhub, OnlyFans and Fansly. Not only could performers lose income, Walters said, but they could also find themselves in offensive or abusive scenes they never consented to.

Lana Smalls, a 23-year-old performer whose videos have been viewed 20 million times on Pornhub, said she has had colleagues show up to shoots with major studios only to be surprised by sweeping AI clauses in their contracts. They had to negotiate new terms on the spot.

In December, the union representing mainstream television actors ended months of negotiation with major studios by approving a new contract that included regulations around the use of AI likenesses. An adult-industry equivalent is unlikely, said Mark Spiegler, an agent representing major names in adult entertainment including Riley Reid and Asa Akira. Spiegler said he gets inquiries about buying or licensing his clients’ AI likenesses multiple times a week.

“This industry is too fragmented for collective bargaining,” Spiegler said. “Plus, this industry doesn’t like rules.”

Inside the conference hall, AI companies dotted the show floor. At the booth for SirenAI, a start-up that makes personified chatbots, performer Jenna Starr talked with the company’s co-founders. If Starr licensed her image and voice, the company could build a Jenna-bot that spits out custom photos, voice notes and messages on demand. Fans could request and receive nude photos of Jenna without any effort on her part.

She peppered the SirenAI team with questions. What cut of her revenue would SirenAI take? Could she terminate her contract at any time? Who would be responsible for promotion?

Then she walked away across the show floor, still thinking. Being on call for subscribers can be a strain on her mental health, she said. Men want to discuss their personal lives. Their requests can be demoralizing: “Pour milk all over yourself. Now drink the milk. Now put on that outfit I bought you last year that you don’t even own anymore.”

Maybe AI will make things better, she said. But she wants to see the numbers first.