To all the new subscribers, welcome. To the everyone who were already subscribed, thanks for sticking around.In 1985 two seasoned Japanese film-makers came together to start a new animation studio and blow a new wind through the world of anime and visual arts. That was the birth of Studio Ghibli.
Forty years later, on March 25 an unassuming Tuesday, OpenAI announced an update to their ChatGPT AI models.
During the demo, Sam Altman and his two employees doing the demo took a selfie of themselves and converted that into “an anime frame”, at least that was the prompt. What ChatGPT spit out looked right out of a dystopian Studio Ghibli movie. A day later, Sam Altman himself, changed his profile picture on X, the platform formerly known as Twitter, to a Ghibli-fied version of his earlier profile picture. He also tweeted that he had changed his profile picture and hoped someone else would make a better version for him.
The result of this whole thing was that people all over the world went and uploaded their photos to ChatGPT and made them into studio Ghibli-style art. People were passing around prompts, news articles were being written about it, and a whole can of worms got opened in terms of artists' rights and ethical use of AI. Everyone behaved like a one-year-old who had their first candy and got really hyper with the sugar high.
Everything was a bit too sweet.
Personally, I don’t follow the world of AI technology too closely. In fact, I had to go find the demo I mentioned earlier on OpenAI’s YouTube channel. What I noticed was that my Reddit and Instagram feeds were getting flooded with Ghibli-fied images in the next two to three weeks.
But something irked me. The kind of feeling that comes with spending too much time in random internet rabbit holes and a strong general suspicion that always makes me say, ‘Something’s fishy’.
The Ghibili-fying of images has become a sign of an advancement in AI technology and a step closer to AGI or Artificial General Intelligence that the AI influencers were wishing for.
And here I was thinking, isn’t that a bit late from a technical point of view? It doesn’t sound like the technological leap that the AI fans claim to be. At least these one-off Ghibli-fied images. I mean, Instagram and Twitter are flooded with some of the most realistic AI-generated ads. Near-perfect-looking AI human beings have already started selling me skin care products.
Sidebar: 'AI human beings' is an oxymoron and if no one has already claimed it to be their invention, I would like to do that, please.
If you go check some corners of the internet, you can find Ghibli-style images made in MidJourney and Stable Diffusion as far as two years ago. Even by OpenAI’s standard, this seems a bit out of pace. I mean, they were showing Sora, their video generation platform, three months ago. So why Ghibli now? After all, all the problems generative AI have with figuring out the number of fingers, the accuracy of facial expressions, and even texts for that matter becomes less of a worry when what we are creating are anime-style illustrations, which by their very nature are exaggerated and doesn’t need to be accurate at all.
Part 1: What the tech?
It is no secret that OpenAI has some of the best people working for them. The technology at its core seemed really impressive and everyone is being blown away by it. For starters, ChatGPT can now make texts without typos. As someone who struggled and continues to struggle with spelling, I always empathised with ChatGPT. It was like the friend who always gave me company at the bottom for all the primary school spelling tests. Now it feels like it has suddenly scored full marks and I feel alone here. ChatGPT images were famous for getting the fingers wrong and the new model is getting better at it. Funnily, the Ghibli-fied image made during the demo got the fingers wrong. A live screw-up, something that the tech demos don’t seem to care much anymore.
The major change is in the way these images are essentially generated. Images are generated by predicting chunks of pixels one by one, similar to how text is generated. A model very different from what they were doing earlier and what everyone has been doing. This is clearly visible as the image is getting generated. Rather than a hazy image getting clearer gradually, the new images on ChatGPT render top-down. Apparently, this is a big deal, at least that’s what the tech analysts and AI Bros are frothing over.
So OpenAI made some impressive technology. Awesome. But somehow the hype was manufactured around Ghibli-fying selfies and memes. And yes I do think it was well-planned hype and not serendipitous in any shape or form.
So, why did they do that?
Part 2: The data draught
In November 2024, a big topic in the AI community was the plateauing of data available to train the AI models. Computer scientists called it, the AI slowdown. Open AI and Meta are pushing for the US government to make copyright non-applicable for training AI models.
So one argument I see on the Internet surrounding the Studio Ghibli hype is the fear that the images you upload are essentially going to be fodder for ChatGPT’s training. After all, users are willingly uploading the data.
This is a serious privacy allegation. Internet watchdogs and even cyber security wing of law enforcement are warning people to be cautious in uploading personal photos to be converted to the Ghibli images.
Does OpenAI use our inputs to train its models? It surely does but the control for it rests with the users. It’s buried deep down in the settings. The user has to make five clicks from the home screen to see the specific toggle in settings to turn it on or off. The switch or toggle is named “Improve the model for everyone”. An altruistic plea would be a generous read. Confusing moral baiting would be a not-so-generous one.
By OpenAI’s privacy policy, this switch, by default, is turned off for premium and enterprise customers and turned on for free users. Hidden, under five clicks, is that inconvenience the cost of using free things?
So in all fairness, OpenAI has an option to turn this off and I don’t think there is any mastermind plan here to steal data. I mean, if they wanted to do that they would have already been doing that. However even if ten per cent of people who used to turn their photos into Ghibli art have the switch turned on, it would be a ‘not so bad’ number. Especially when the data wells are drying out and getting a diverse set of images, especially people with different skin colours and features, would be really nice.
Part 3: Agents at the door
Up till the end of March, the conversation around AI was shifting more towards AI agents. Everyone who was developing AI Agents had a different definition of what an AI Agent was but one thing they all agreed on was that the target market is not individuals but enterprises.
With this shift towards Agents and enterprise solutions, the competition started increasing and the edge OpenAI had built up slowly started to get dull.
The Ghibli trend essentially brought back the conversations around AI to everyday consumers. It makes sense when you realise that the lines of division drawn three or four generations ago, between consumer and enterprise application has become irrelevant now.
Imagine someone high up in the hierarchal ladder in an organisation. If they trust a company with their personal information, selfies and pictures of their family, then vouching for that company’s API become much simpler. Almost subconscious.
The fight here is for a slice of our trust. This could be why OpenAI is willing to “melt” their GPUs and bleed money.
Part 4: The grand-prix
On March 25, the day OpenAI demoed its new image generation tech, Google made an announcement. “Today we’re introducing Gemini 2.5, our most intelligent AI model.”
‘ChatGPT feels like a joke compared to Gemini 2.5’, a fairy viral post on the subreddit r/Gemini. More unbiased sets of sources said Gemini 2.5 was really competitive with ChatGPT, even performing better in deep research and reasoning. But most importantly, Gemini is free to use. Free AI, in this economy?
This race to beat the competition on the calendar is not something new. Tech reporters like Casey Newton, with their inside information even believe that there was even a race to the starting line.
He says that the launch of ChatGPT was postponed because Anthropic was about to release Claude and Sam Altman wanted to be there first.
Then there is Mira Murati. The ex-CTO of open AI, who started Thinking Machines Lab in February of this year and is already aiming at raising 2 billion dollars, billion with capital B, in seed funding. There are countless others like Ilya Sutskever or Andrej Karpathy who left open AI to start their own AI companies like Safe Superintelligence and Eureka Labs.
Reminds me of this scene from Futurama, which has become a meme now.
Another pivotal moment in this short history was China’s late entry into the track. On January 27, the Deep Seek chatbot was launched and it hit everyone in Silicon Valley like a giant autonomous truck. It soon overtook ChatGPT as the most downloaded freeware app on the iOS App Store in the United States.
There are many now. Grok, Gemini, Claude, Safe AI and a hundred others. Even MS Paint, one of the best pieces of software ever, has AI. What’s happening is that the pie called the market was getting shared faster than the size at which it was growing.
ChatGPT launched with record-breaking success in late 2023. They surpassed 1 million users in just five days. Then it was all uphill, until Jan 2025 and they saw their first dip of 2.8% in the total people who visited ChatGPT.
One response to this has been Sam Altman’s constant ass-kissing towards India. When the whole Ghibli trend was going on, Sam Altman tweeted/re-tweeted a total of 5 Ghibli-fied images, including the one from the demo live stream. Out of this, one was a repost of Ghibli Modi (a phrase I never thought I would say when I woke up today), another repost of Ghibli images from India Post and finally, Ghibli Sam Altman playing cricket in the Indian cricket team’s Jersey. A leaf right out of a Western influencer playbook. Want to increase subscribers or followers? Want to hide a controversy in a pile of engagements? Visit India and just pretend like you are enjoying it.
Sidebar: Is that what they think of us Indians? Modi, Cricket and Outdated forms of communication? Oh no.
Maybe I am reading too much into it. But whatever it is, Open AI saw a million people signing up for ChatGPT in an hour to convert their photos into Ghibli style.
Part 5: The Inescapable grip of Internet virality
The essence of this rant is that, there is no escape from the modern-day algorithmic virality even if you are the hot new technology in town. I think Sam Altman knows that and this is not the first time he has tactfully used a moral controversy to push new features. A year ago in May 2024, Sam Altman tweeted ‘Her’. Within a few days, ChatGPT released its voice assistant feature with the voice of Scarlet Johansen. Her is this famous movie where Scarlet Johansen does voice for a creepy AI. Scarlet Johansen never said okay to OpenAI using her voice. What followed was huge public controversy and ChatGPT again became a viral topic. With the Ghibli trend as well, it was something similar. A lot of posts on the ethical and moral complexity surrounding it and a lot more on copyrights.
I mean this is a controversy that OpenAI can clearly manage compared to the controversy surrounding how AI Agents are going to make tons of jobs obsolete, the amount of energy AI is using and the associated environmental impacts. This is a virality that is more legal in nature and that they seem to have some sort of control over. This virality also gave OpenAI a cover to move closer towards the right of the political spectrum. Guardrails of content moderation were being taken down in the name of free speech. What we saw happen with X, the platform formerly known as Twitter. Then Instagram followed suit and now ChatGPT. This according to me has to be the biggest change that was announced as part of this particular product update. Ghibli-fied pictures of Hitler in front of a Paris, historic riots and Dictators started popping up. A bit closer to home, the infamous photo of the Babri masjid being destroyed by Hindutva mobs, an event that made some of the ugly stains in Indian history was being Ghibli-fied and posted on twitter and Instagram.
Conclusion: So what?
I think what OpenAI did with the Ghibli update was not a technological leap, but rather a clever UX innovation. People had to use complex prompts or connect various AI models locally to make these images. With this update, it can be done with ChatGPT's minimal interface, along with the rest of the user's conversations.
The Ghibli trend was extremely impactful. We were this close to discussing AI misinformation, we were this close to discussions on super-intelligent AI agents taking away jobs, we were this close to talking about a probable AI bubble that was about to burst. A sweeping stroke of Japanese-style anime and like the contemporary philosopher Olivia Rodrigues said, “It's always one step forward and three steps back”.
I think this mass Ghibli-fication is going to be a pivotal point in the history of AI. Not in technology but rather in the experience of using AI by everyone. A case study on increasing user adoption of technology by appealing to an insatiable need to be bedazzled.
Before I wrap this up. I don’t have many moral qualms about Ghibli-fication. I was not cool enough to know Studio Ghibli movies, as a kid from a village in some out-of-shape corner of Kerala. So it didn’t define my childhood or anything.
I didn’t Ghibli-fy myself because I am paranoid but more importantly I just did not like it.
Like I said, a bit too sweet for my palate.
I understand. The world’s on fire. Literally and figuratively. It feels like we ate a handful of bird’s eye chilly and downed it with a bottle of hot sauce.
Maybe these sugary, saccharine AI images of us and our world are what we need. At least for some temporary relief.
If all this sounds like very tinfoil-hat-conspiracy stuff to you, would your perspective change if I call it an Opinion piece?
https://ghibli.geoarthouse.com
https://www.theverge.com/openai/636529/images-chatgpt-openai-studio-ghibli-copyright
https://open.spotify.com/episode/6orAhWCiIuoo9I3qgXhMcL?si=ff017e67e0404465
They say, if you want to see the root of any opinion or research, you have go deep and find who is funding it. I am not an exception to that. I am being funded by me so if you feel like supporting, please consider subscribing and share this with someone who might want something to go to sleep.
Suggestion Corner
You play *as* the anxiety.
Share this post