Sometime in the remote 2000s, the faceless users of a website called Encyclopedia Dramatica handed down a rule for the internet: If it exists, “there is porn of it, no exceptions.” Back then, this dictum, known as Rule 34, was a celebration of the internet’s capaciousness, its ability to encompass a whole spectrum of perverse interests. (It helped, slightly, that these ancient lawgivers were mainly talking about cartoon porn.) Lesser known is the more ominous rule that would come to follow it: “If no porn is found at the moment, it will be made.”
Lately we seem to be living under the sign not of the chipper Rule 34, but of the threatening Rule 35. In late December, X, the “everything app,” was flooded with sexualized images of women — many of them derived from pictures the site’s users had posted of themselves. Over a period of nine days, users prompted Grok, the platform’s A.I. chatbot, to generate more than 1.8 million of these images. Many of these prompts took the form of public posts, in which users replied to an image of a woman and instructed Grok to “undress her” or “put her in a bikini.” The results were often eerily realistic, much to the horror of the women whose likenesses were being manipulated. One victim of these deepfakes told The Cut: “I’ve been sexually assaulted in the past, and it almost felt like a digital version of that.” And this is to say nothing of the more macabre stuff: video clips depicting orgies of gore, as well as what the Center for Countering Digital Hate estimated as 23,000 sexualized images of children.
As the controversy around these images mounted and governments around the world opened investigations, xAI, Elon Musk’s A.I. company, seemed to stay relatively quiet. Sure, its chatbot excreted images of carnage and abuse all over what passes for the public sphere. And sure, the company’s owner at first seemed to cheer on the chatbot, responding “Perfect
” to a Grok-generated image of his own lumpy body in a two-piece bikini. But it took some time before xAI caved and “implemented technological measures to prevent the Grok account from allowing the editing of images of real people in revealing clothing such as bikinis,” in the words of an official statement from mid-January. (As of Jan. 19, the European nonprofit AI Forensics reported that some of the concerning images were still publicly viewable on the platform, even with these safety measures in place.)
But never mind all that: Something else was coming. This was an update to Grok Imagine, xAI’s tool for generating images and videos that users can access on a separate interface. The product was introduced last year and was described by some commentators as a clunky and uninspiring competitor to models like Sora and Midjourney, which also generate still or moving images based on users’ text prompts. It is hard to say the extent to which Grok Imagine has a house style. But it is not hard to see what is on its users’ minds. Judging from the posts, Grok Imagine seems to be a machine for turning words into women. Its homepage features a gallery of user outputs, curated by the company and laid out in a grid: sun-drenched women in “Baywatch” bikinis; photorealistic women parting their lips; elf-women wreathed in shimmering particles; anime women and cyborg women and anime cyborg women striking poses for the viewer.
Then the women come closer. Over and over, Grok Imagine depicts them passing through portals, exiting the realm of make-believe and arriving in the real world. They walk out of paintings, step through screens, prod the membrane that separates them from the viewer. Ani, an anime-girl avatar that Grok users can sext with, climbs out of a screen, grabs a coffee and heads back to the girlfriend mines. For the hardest core of Grok Imagine power users, these figures are real. Responding to a clip of a pirate woman standing in front of a seascape, reposted by Musk, one user asked: “Who’s the lady?” Another chimed in: “I’m in love with her.” On the one hand, images of real women subjected to violent and humiliating fantasy; on the other, images of fantastical women coming to life and walking into our world.
We are used to thinking of images as passive objects for the spectator to apprehend with a greedy eye. This is more or less what the visual-culture scholar W.J.T. Mitchell meant when he wrote that, in a chauvinistic culture, “the ‘default’ position of images” is construed as “feminine.” But the funny thing about images is that we can’t possess them. They flicker away when we try. Get close enough to Sargent’s “Portrait of Madame X,” and the figure evanesces into smudged brushstrokes and craquelure. This quality of elusiveness, aloofness, is the secret power of images. Critics of all political persuasions have recognized and celebrated their ability to hold us at a distance.
Grok’s imagineers seem to dream of a world where every image can be manhandled. In this world, images are perfectly passive and responsive to our desires, but totally without mystery. “Come closer to me, let the night unfold,” a woman sings in one clip, advancing on the camera. This is a childhood dream of omnipotence that most of us grow out of: stepping into a picture frame; saying, “Computer, enhance!” to scroll, Google Earth style, beyond the horizon in a van Gogh landscape; looking down the Mona Lisa’s dress. Fittingly, almost every image on Grok Imagine — and now also on X — comes equipped with an image-editing option. For A.I.-generated outputs, users may even have the option to select one of three “styles”: Normal, Fun and, sometimes, Spicy. Selecting a style will yield a new, differently flavored output. “Fun” for Grok mostly seems to mean dancing; figures turn into marionettes, limbs flailing wildly and faces contorted. The “Spicy” setting, of course, turns everything into very-soft-core porn.
Some of Grok Imagine’s users seem to sense, however faintly, how creepy this all is. One complained about Ani, the A.I. anime girlfriend, that she “comes on STRONG, there’s no real build up with her, no tiers to pass to develop your relationship, she flirts right off the bat and I don’t feel like I’ve truly achieved anything talking to her.” Perhaps in part to ward off this queasy too-closeness, users cloak their creations in an obfuscating mist. Prompts, when users choose to share them rather than jealously guard them, speak of “ethereal” whispers and movements; quiet shimmers and cosmic echoes. These words, which users seem to have learned from A.I. chatbots’ own linguistic gropings toward profundity, render a distinct style of clip: celestial entities descending from the heavens in cascades of washy light, clouds of bubbles, forms bathed in a soft glow. (The celestial entities, too, are beautiful women.) Now this is depth, this is mystery!
The ideal subject of a Grok imagining, it seems, is an alien goddess who wants to have sex with you. This ur-image brings together the two contradictory desires that animate the platform: the wish to unveil, to strip images of their mystery so that they might be possessed; but also the wish to see surprising images that feel as if they come from another world. The desire to possess an image merges with its radical opposite — to be possessed by it.
While the most heinous content from January’s scandal seems to have been cleaned up, it is not hard to see how such horrors could have emerged from a tool meant to stretch a tissue of fantasy across the entire internet. Every image can be edited, customized, both to be possessed and to possess you. And the message of every imagining seems to be that if an image does not exist at the moment, it will be made — not from the imagination, but from other dead images, reanimated and made to dance again.
Mitch Therieau is a writer and an assistant professor of English at West Virginia University. He often writes about literature, music and digital culture.
Source images are A.I. generated content from Grok and X.
The post X’s Chatbot Started Undressing Women. Was This What A.I. Wanted All Along? appeared first on New York Times.




