For the past couple of years, a certain type of animation has haunted me. It’s an AI-manipulated image that brings its human subjects to life by simulating speech, lip movements, and facial expressions. In these ‘talking head animations’ or ‘digital puppets,’ the background remains static while only the subject’s face moves or changes. The result is an image caught between stillness and motion—an ‘in-between image’ that exists across multiple temporal dimensions

Even though I haven’t found a single article about them, I keep seeing them everywhere: in true crime TikToks, news reporting, and in artworks by the likes of Jon Rafman, Felix Burger, Sean Capone, and Poorspigga. They even showed up in a Netflix documentary I watched recently, Dirty Pop—a film about Lou Pearlman, the music mogul behind NSYNC and the Backstreet Boys. In this documentary, since Pearlman was not alive to recount the story himself (he died in 2016), the producers created a talking head animation of him, using his own words from his 2003 biography. It is therefore not Pearlman who is speaking in the documentary, but his AI-resurrected avatar.

The use of talking head animation to resurrect the dead goes far beyond the case of Pearlman. For example, the app DeepNostalgia, created by the genealogy platform MyHeritage, allows users to animate old photographs of deceased ancestors—making them smile, wink, or even blow a kiss. A darker and more unsettling form of resurrection appears in a true-crime TikTok trend, where fictional AI-generated children narrate the stories of their own deaths. In one of these videos, an AI-generated two-year-old child with big brown eyes looks into the camera and says, “My grandmother stabbed me and put me in the oven alive. I was only 20 months old.”

@hueygspy794 I was put in the oven by my grandmother!#tiktok #fyp #story #suffering ♬ original sound – My life

 While the use of this type of image, as in the case of Pearlman, might spark a number of controversies—such as “Who gets to speak for the dead?”— there is something seemingly more benign about these kinds of animations compared to the perhaps more widely talked about deep fakes. While talking head animations and deepfakes both involve manipulating footage to make it appear as though a person is speaking or moving in ways they did not, they differ fundamentally in intent and execution. Talking head animations focus solely on animating facial movements without much potential to deceive, whereas the primary aim of deepfakes is precisely to create a convincing illusion, often with deceptive purposes. For the rest of this text, I will therefore discuss these animations not in terms of ethics, but in relation to time, memory, and Derrida’s concept of hauntology.

While photographs and videos are typically associated with two key points in time—when they are created and when they are viewed— talking head animations introduce a third point: when the image is animated. It can be useful, then, to examine the relationship between these first two moments and consider how this influences our perception of the animated image.

To understand this relationship, take John Rafman’s project Punctured sky, which consists of a video work as well as a series of TikToks – both featuring talking head animations. The “image world” that Rafman animates consists either of real, found footage of people from the 2000s early-internet era or AI-generated to evoke the same aesthetics, with signifiers such as LAN parties, bulky computers, Alienware rigs, Windows XP, CRT monitors, and video rental stores. While the images themselves are frozen in the early 2000s, Rafman “resurrects” them using AI in two key ways: by applying animal AR filters to the characters’ faces, blending human and animal features, and by animating their faces and mouths so that the images “talk.” 

The “high”-tech, 2020s AI aesthetic is therefore juxtaposed with the low-tech visuals of the 2000s, creating a jarring, distorted and uncanny new image. This relationship between the past and the present evokes Derrida’s concept of hauntology, which articulates how the present is always haunted by the ghosts of the past and the unfulfilled promises of the future. The unfulfilled promise in this case is how, in the early 2000s, the internet was widely perceived as a democratizing force, a space that one expected freedom, decentralization, and an egalitarian future. The imagery in Rafman’s work from this period—bulky computers, physical video stores, and the communal spirit of LAN parties—captures this optimistic outlook.

However, this utopian vision of the internet never really materialized. Instead of becoming the decentralized, liberating space many hoped for, the internet has evolved into something quite different: a centralized, commodified, and heavily surveilled environment dominated by powerful platforms and corporations. Watching these animations, with their unsettling hybrid appearances and mechanical movements, forces us therefore to confront the gap between the past’s idealism and the present’s disillusionments. They haunt us with the lost futures of the internet.

This hauntology is present not only in the imagery used by Rafman but also in the storytelling of Punctured Sky. The short film follows a man who remembers playing a game from his childhood, but despite his best efforts, he cannot find any record of its existence. Here again, the past and its ghostly absence haunt the main character, making his quest both a search for a lost piece of his history and a confrontation with the unsettling void left by the unfulfilled promises of the early internet. The title, PuncturedSky, further reflects this theme, suggesting a disruption in a once-anticipated future.

It seems like Rafman’s character is not the only one grappling with nostalgia and loss for the promises of the early internet era. Reflecting this sentiment is the Time Traveler TikTok trend, which features AI edits of viral videos from the late 2000s and early 2010s, such as Charlie Bit My Finger and 9+10 kid. Initially, these videos present familiar, nostalgic content, but they soon morph into disturbing AI-generated versions with grotesque distortions and an unsettling soundtrack. In these videos, the “AI time travelers” often attempt to alter or erase the past. For example, in the Charlie Bit My Finger video, they intervene to stop Charlie from biting the finger, while in the 9+10 video, they prevent him from saying the iconic phrase “21.” The role of these time-travelers seems, therefore, to rewrite the more benign earlier internet history in order to naturalize the current, dystopian state of the internet in an act of historical revisionism. 

It makes sense why these images with AI-interventions have haunted me so much. Or rather, it is not the images themselves that have haunted me, but the lost futures they carry with them—the echoes of a digital utopia that never came. As one commenter noted on a Time Traveler video, ‘Bro is making me rethink my childhood😱.’

@milkman8008135s @milkMAN 🤯 no time traveler stops charlie from biting his finger‼️ #ai #vine #timetravel #aimemes #meme #charlie #fyp #foryou #viralvideo ##timetraveler ♬ MARKING TIME, WAITING FOR DEATH – Shiro SAGISU