Arya News - Millions are uploading selfies for a retro AI makeover. Almost nobody asks where the photo goes or what it took to generate. The 80s aesthetic will fade from feeds within weeks. The data does not fade on the same schedule, and neither does the energy and water it took to make it.
NEW DELHI – Open Instagram or X (Twitter) this week and you will see it everywhere. Big teased hair. Neon lighting. Mall photo studio backdrops. Millions of people are uploading a current selfie to ChatGPT and asking it to turn them into a version of themselves from 1986. It looks like harmless fun. It has already spread across South Asia, the US, and the Middle East, and Google Trends listed it as a top search this week. But the trend hides two costs that almost nobody talks about while they post the result. One is what happens to your face once you hand it over. The other is what it takes to generate that image in the first place.
How the trend actually works
The format is simple. You pick a clear photo of yourself, upload it to ChatGPT, and give it a prompt describing an 80s look. Popular versions include the yearbook portrait, with teased hair and a mottled blue studio backdrop, and the neon synthwave portrait, with pink and blue lighting and a leather jacket. The tool keeps your face recognisable and rebuilds everything else, the hair, the clothes, the lighting, the grain. That is what makes it feel so convincing compared to older filter trends, which just added sepia tone or fake scratches on top of the same photo.
The polish is real. Facial consistency in image models has improved a lot over the past two years, which is why this version of the trend spread faster than earlier ones. But that same technical leap is exactly what raises the stakes for the person uploading the photo.
Your face is not just a filter input
When you upload a photo to ChatGPT, you are not just requesting an edit. You are handing over a high resolution image of your face, and depending on your account settings, that image can be used to help train future versions of the model. Cybersecurity researchers have pointed out that a single upload carries far more than the visible picture. It can include facial geometry, skin tone, background details, location clues, and sometimes documents or other people who never agreed to be part of the exchange.
Several experts have raised the same concern in almost identical language over the past few months, as similar viral photo trends kept resurfacing. A cybersecurity researcher at the University of Alabama at Birmingham explained that the model studies more than the picture itself. It can pick up biometric details like eye colour and hair colour, information that has value well beyond a fun retro filter. A researcher at Counterpoint Research described the pattern more bluntly, calling these trends a steady stream of free biometric data that keeps feeding AI systems long after the trend itself fades from people’s feeds.
The bigger risk shows up if that data is ever mishandled. One security expert warned that leaked images containing detailed facial features could later be reused for fraud, blackmail, or impersonation, and once that data escapes, there is no way to fully take it back. Deepfake risk and identity theft are not hypothetical concerns tacked onto a scare headline. They are the direct, practical consequence of large numbers of people uploading clear, high quality face photos to the same handful of platforms in a short window of time.
To be fair, this is manageable if you take a few precautions. Turning off model training permissions in your account settings, avoiding photos that show documents or other identifiable people, and stripping location metadata before uploading all reduce the exposure. Almost nobody does this before joining a viral trend, though. The whole appeal is speed and spontaneity, which is precisely what data protection requires you to slow down for.
The environmental side nobody posts about
Every one of these images has to be generated somewhere, and that somewhere is a data center running on electricity and water. Estimates of the energy cost per image vary widely depending on the model and settings, but even the low end adds up fast at the scale of a viral trend. Research from Hugging Face and Carnegie Mellon University found that a single image generation can use around 0.01 kilowatt hours of energy, roughly half of a smartphone battery charge. Other studies measuring newer diffusion models found energy use per image ranging from a fraction of that up to several times higher, depending on resolution and the number of processing steps involved.
Multiply that by the scale of a trend spreading across three continents in a matter of days and the numbers stop being trivial. Stability AI reported that users generated more than 12.7 billion images on its models in a single year. At an average of even a few thousandths of a kilowatt hour per image, that scale alone adds up to tens of gigawatt hours of electricity, comparable to the annual usage of thousands of homes.
That electricity does not come free of water either. Data centers rely heavily on cooling systems, many of which use evaporative or “swamp” cooling that consumes fresh water directly. The International Energy Agency has estimated that global data centers use hundreds of billions of liters of water a year for cooling, and a large share of new AI infrastructure is being built in regions already under water stress. A single 100 megawatt data center in the US can draw around 2 million liters of water a day, roughly the equivalent of 6,500 households.
None of this means one selfie is destroying the planet. It means a trend built on hundreds of millions of individual generations, repeated across yearbook shots, synthwave shots, and endless retries to get the hair right, is quietly adding up on infrastructure that is already strained in places that can least afford it.
A trend that deserves more skepticism, not less
This trend is not the harmless novelty it presents itself as. A filter used to mean changing colour tones on your own device, with nothing leaving your phone. This version means sending a clear image of your face to a company’s servers, feeding a model you cannot inspect or audit, and drawing on electricity and water infrastructure that is already strained in some of the places it operates. That is a much bigger ask than a photo trend usually makes, and most people are agreeing to it without realising they were asked anything at all.
The privacy risk is not evenly distributed either. Once biometric data leaves your hands, you have no real way to verify it was deleted, and no way to control what a future breach or policy change does with it. The environmental cost is not something any individual chose either. It was built into the infrastructure before the trend existed, and it scales up automatically every time millions of people repeat the same request for a slightly better version of their neon backdrop.
None of that gets weighed against the fifteen seconds of nostalgia the trend delivers. A viral format spreads because it is fast and fun, not because anyone stopped to ask what it costs. That gap, between how quickly something spreads and how little scrutiny it gets, is the actual story here. The 80s aesthetic will fade from feeds within weeks. The data does not fade on the same schedule, and neither does the energy and water it took to make it.