#e4mXplains:  Maybe don’t always upload your personal photos onto image generation tools

The internet loves a trend. From the heart-warming, if body-chilling, Ice Bucket Challenge, to the corporate debacle that was Tide Pods, we do love getting involved.

Over the past year, that has meant a succession of AI image trends: first turning ourselves into Studio Ghibli characters, then into Nano Banana figurines and retro portraits, and now apparently into inhabitants of the 1990s. 

Each one follows roughly the same ritual: find a decent photograph of yourself, upload it to an image generator, give it a prompt, and wait to see what the machine thinks you should have looked like in another universe.

Read On: Could advertising data become a national-security risk in India?

And the appeal is obvious. Most of us are never going to star in a period film, pose for a vintage magazine cover, or be lovingly hand-drawn by a Japanese animation studio. An image generator gets us reasonably close in the time it takes to upload a photograph and type a sentence.

And because the output is us, only shinier, stranger or more cinematic, it is almost engineered for sharing. One person's 1990s portrait becomes five others'. Someone turns themselves into a figurine, half their office follows, and by lunchtime LinkedIn looks like a very expensive toy shop.

This is, mostly, harmless fun. It is also a very effective demonstration of how quickly AI image generation has become ordinary. Not something done by designers, VFX artists or people with alarming graphics cards, but by anyone with a phone, a face and a few idle minutes.

The interesting part is what happens once handing over that face starts to feel completely unremarkable.

Read On: The data you don't have is killing you

The Digital Wild West

Ever since we scaled the ability to broadcast ourselves to the world, through internet forums and Hi-5s to Instagram pictures and YouTube Lives, we've been putting ourselves, and our physical information, out there. And because all that information was sitting out there, it was used. In various ways, none particularly pleasant.

India has already had its introduction to what happens when a likeness becomes malleable. Celebrity deepfakes have been used for everything from fake investment endorsements to pornography and impersonation. 

A 2026 McAfee survey found Indians reported encountering an average of four deepfakes a day, while one in five respondents said they had encountered a voice-clone scam. More than a third said they were not confident they could identify a deepfake scam. These are survey figures rather than police records, but they give some sense of how ordinary synthetic media has become.

The broader Indian cybercrime numbers provide the less glamorous backdrop. According to data tabled by the Ministry of Home Affairs, the National Cyber Crime Reporting Portal received more than 24 lakh financial-fraud complaints in 2025, involving over Rs 22,495 crore in reported losses. Those are not deepfake-specific figures, but they show the ecosystem into which increasingly convincing impersonation tools are arriving.

Read On: Meta-Cred deal sparks data ownership & privacy conversations

And images themselves can now do significantly more damage than merely embarrass their subjects. Globally, the Internet Watch Foundation identified 8,029 realistic AI-generated child sexual abuse images and videos in 2025, while the US National Center for Missing & Exploited Children received more than 182,000 reports involving the possession, generation or attempted generation of AI-generated CSAM that year. NCMEC also recorded tens of thousands of cases in which users attempted to create such material by uploading an existing file and prompting an AI system to alter it.

India has consequently begun writing synthetic media directly into its internet rules. Amendments to the IT Rules notified in February 2026 specifically address synthetically generated information, including deepfakes, and require stronger platform safeguards around unlawful AI-generated material, including impersonation, non-consensual intimate imagery and child sexual exploitation material. Sensitive complaints involving nudity, sexual content or impersonation are subject to a two-hour grievance-response timeline.

Modern Living

This wasn't our fault. Bad actors will always find a medium. 

We sometimes forget how quickly all these capabilities came around, from the first pixelated selfies, and how quickly they advanced, to the point where you can come, in glorious 4k-resolution, on to the screens of tens of millions of people around the world at the click of a single button. Or to the twos of people. As with much else, it depends on your social media clout.

And so we happily consent to what can so easily become the invasion of privacy. It should also be noted of the growing cases of outrage amongst Gen Alpha and young people growing up, many of whom are not happy at their doting parents putting up video evidence of their childish shenanigans for the edification of Facebook friends and internet strangers.

In any case, we're consenting again. To another kind of invasive privacy. Perhaps not a public one, but definitely a very mutable one.

Read On: Robot Anchors and Phony Photos: India Confronts AI-Generated Content

In March 2025, when OpenAI introduced native image generation into ChatGPT, social media feeds were promptly Ghibli-fied. People uploaded photographs of themselves, partners, children and pets to be redrawn in the visual language associated with Studio Ghibli. OpenAI says more than 130 million people generated over 700 million images in the feature's first week. Amid the frenzy, company COO Brad Lightcap described India as ChatGPT's fastest-growing market.

Five months later came Google's Nano Banana, the nickname for its Gemini 2.5 Flash Image model. This time Indians turned themselves into tiny 3D figurines, old Bollywood stars and, memorably, women in 1990s-style saree portraits. By September 12, 2025, Google said more than 500 million Nano Banana images had been generated worldwide and India was among the feature's leading markets. Nano Banana trends subsequently made Google's own Year in Search review for India.

Now we have moved on to another round of retro portraits. The aesthetic changes. The underlying instruction increasingly does not: upload a good photograph of yourself.

There is an important distinction here. Uploading a photograph to an AI image generator does not mean that photograph is automatically dumped into some giant training folder with your name written on it. What happens depends on the product, the account and the privacy settings being used.

OpenAI says content submitted through its consumer services may be used to train and improve its models. That can include uploaded files and images. Users can turn off “Improve the model for everyone”, after which new conversations are not used for training; Temporary Chats are also excluded. OpenAI says it takes steps to reduce personal information in datasets used for training.

Google is more explicit about Gemini. For adults, its Keep Activity setting is on by default. When it is on, chats and material shared with Gemini (including photographs and files) can be used to improve Google services, including training generative AI models, with some material reviewed by humans. Turning Keep Activity off prevents future chats from being used to train Google's AI models unless the user subsequently submits the interaction as feedback.

None of this is illegal, hidden in a basement, or necessarily even objectionable. Models improve partly by learning from how people actually use them, and both companies give users controls over that process.

But consider what an AI-image trend produces. A user supplies a photograph, explains what they want done to it, sees the result, asks for changes and perhaps tries several times until the machine produces something worth putting on Instagram. 

Read On: Instagram expands marketing API with new metrics and content management tools

Where the user's settings permit that interaction to be used for model improvement, they haven't simply supplied a picture. They have generated an unusually rich little bundle of information about a face, a desired transformation and what they considered a satisfactory result.

Multiply that behaviour by the hundreds of millions of images produced during Ghibli, Nano Banana and whatever comes after the current retro craze, and the silly internet trend begins to look rather more interesting.

And for users who have agreed to let their data be used to improve the model, there is another layer to all this. Every uploaded photograph, every prompt asking for a Ghibli face or a 1990s portrait, every regeneration and adjustment can become useful material for improving how the system handles image generation. In other words, while the model is making pictures for us, some of us are also helping teach it how to make better pictures of people like us. 

Again, there is nothing inherently sinister about that. But it is worth recognising that the fun little trend is also, potentially, a very large and very willing training exercise.

The Old Advertising Bargain

Advertising has, of course, been putting a price on our privacy for decades. The modern digital advertising industry was built in considerable part on turning things people did online (searches, clicks, purchases, locations, viewing habits) into signals valuable enough for somebody to pay to reach them. In return, consumers received an internet filled with services that were free, cheap, convenient or personalised enough that the bargain generally seemed worthwhile.

Generative AI adds an unusually intimate layer to that transaction. We aren't merely telling a platform that we looked at shoes or watched three cooking videos. Increasingly, we are voluntarily handing over our faces, voices and creative preferences, often accompanied by very precise instructions about what we would like those things to become.

That doesn't mean the selfie you uploaded to make yourself look like a 1990s film star is being used to sell you shampoo. The point is broader: personal information becomes economically useful because it makes digital products better at understanding, serving and retaining their users. And where brands begin adopting the same mechanics of upload your photograph to enter our campaign, virtually try on a product, put yourself inside our ad, privacy becomes a marketing question as much as a technology one.

Which makes consent slightly more complicated than clicking “I agree”. India's new data-protection regime is moving toward requiring companies to tell users, in plain language, what personal data they are collecting and why. That is important. 

But no regulation can answer the more mundane question for us: how much privacy are we personally prepared to exchange for something sufficiently useful, entertaining or flattering?

Read On: We refuse to let India become a passive consumer of foreign AI tools: Dr Sandeep Goyal 

TL;DR

Privacy has always had a price. For much of the internet age, we paid it in cookies, search histories and the knowledge that somewhere, someone knew we had spent 20 minutes looking at air fryers.

AI simply makes the transaction more personal. The thing being offered in exchange is better too. Not a targeted banner ad, but you as a Studio Ghibli character, a 1990s movie star, or whatever aesthetic the internet collectively decides on next Tuesday.

None of which means you shouldn't do it. It might just be worth remembering that when something online asks for your face before giving you something delightful in return, the delightful thing isn't necessarily the only valuable part of the exchange.

So by all means, make the picture.

Maybe just don't upload every picture.