For a few weeks, it felt like everyone online had been redrawn as a soft, hand-painted character with big eyes and a wind-swept meadow behind them. The Ghibli-style AI portrait trend was fun, fast, and oddly moving — until people started noticing that not everyone came out looking like themselves. This is the story of the craze, the quiet bias baked into many of these tools, and why fair, identity-preserving portraits are worth fighting for.
AI portrait filters have been creeping into our feeds for years. First came the "age progression" selfies, then the oil-painting avatars, then the 3D figurine renders that made your LinkedIn photo look like a collectible toy. Each wave promised the same delightful twist: upload a selfie, get back a version of you reimagined by a machine.
What changed recently was speed and scale. Models got good enough that a single tap could produce a result that looked genuinely artistic rather than janky. When a tool feels that effortless, it stops being a novelty and becomes a habit — your profile picture, your group-chat avatar, your conference badge. The portrait filter quietly became the front door to consumer AI for millions of people who had never touched a prompt.
That mass adoption is exactly why the details matter. A filter you use once for a laugh is one thing. A filter that quietly decides what "you" should look like — and then shows that version to the world — is something else entirely.
Studio Ghibli's hand-drawn worlds have a recognizable softness: warm pastels, gentle light, rounded features, and a deep tenderness toward ordinary moments. For decades, fans described the look as "comforting" and "human" — a style that feels less like a screenshot and more like a memory. It is beloved precisely because it treats small, everyday people as worthy of beauty.
So when image models learned to mimic that painterly softness, the result spread fast. People flooded social feeds with portraits of themselves and their families reimagined in what everyone started calling "Ghibli style." The trend was a love letter to a beloved art form — and, for many, a rare chance to see themselves rendered with the same care the studio gives its characters.
The delight was real. But the same viral moment exposed a problem that fairness researchers had been warning about for a long time: the models producing these portraits had been trained mostly on images of light-skinned, East Asian, and Western faces, and they carried those defaults into everyone's photo.
The uncomfortable part of the trend: for a lot of users, the "Ghibli-style" output didn't just restyle them — it quietly edited who they were. The portrait looked less like them and more like a generic, lighter, narrower version of a person.
Here is the part most marketing pages skip. A large share of popular AI portrait apps do not merely apply a style — they normalize the face underneath it. The bias is rarely a single dramatic change; it is a stack of small, cumulative edits that add up to a different person.
The most common issue is skin-tone lightening. On darker faces, the model often lifts complexion toward a paler mean, smoothing out the very features that make someone recognizable. It is rarely labeled, never asked for, and almost never mentioned in the cheerful "look how cute!" share screen. Users who point it out are told it is "just the style" — but a style should change the rendering, not the identity.
The second issue is shape. Eye shape, nose width, lip fullness, and facial structure get nudged toward a single "standard" ideal that the training data over-represents. The result is a portrait that reads as more Eurocentric than the person who uploaded it. Again, this is sold as a neutral artistic choice when it is anything but — it is the engine of the app deciding which faces count as "normal" to paint.
None of this is a conspiracy. It is the predictable output of models trained on biased data and shipped without anyone checking how they treat the full range of human faces. But the harm is the same whether or not it was intentional: people are told, implicitly and constantly, that the most flattering version of them is a lighter, narrower one.
A selfie is not a defect waiting for an algorithm to correct. It is you. When a tool quietly lightens or reshapes a face, it sends a message — to the person in the photo and to everyone who sees it — that the original was somehow not quite right. Multiply that across millions of portraits and you get a culture where only one kind of face is treated as worthy of the soft, beautiful, "Ghibli" treatment.
The danger is that these defaults are invisible. Most users never see the un-biased version; they only see the output and assume that is what the style "does." That makes the bias self-reinforcing: the more people share whitened, warped portraits, the more the look becomes the expected one, and the more pressure there is to conform to it. Fair representation is not a nice-to-have add-on. It is the difference between a tool that celebrates people and one that quietly erases them.
This is also why we think the conversation should move past "which filter is prettiest" to "which tool respects you." If you want to dig deeper into how this plays out across avatar apps, we wrote about it in our piece on why fair AI avatars matter for non-white faces, and on the privacy side in our guide to keeping your AI avatar private.
At LumTale, we built our own portrait style — Lumling — specifically so we would never have to copy another artist or inherit someone else's biased defaults. Lumling is a new, warm, stylized look we designed in-house: soft shapes, gentle color, a slightly collectible feel. But the style is only half the story. The other half is how we treat your face.
Our generation is built to be identity-preserving. That is a deliberate engineering choice, not a slogan. Lumling changes the rendering — line, color, texture, lighting — while keeping the structural features that make you you intact. We do not whiten skin tones, and we do not warp facial structure toward a single "standard" face. Your features stay faithfully yours, whether your portrait ends up dreamy, toony, or studio-polished.
If you want the technical and ethical details, our About page lays out exactly what we do and what we refuse to do. We also keep our pricing simple and honest — no dark patterns, no surprise training clauses buried in fine print.
Fairness is not only about the output; it is about consent. Most apps are vague about what happens to your uploaded photo. LumTale is not. When you use our cloud studio, your image goes to a vetted GPU partner to render your Lumling portrait — and we tell you that plainly, before you upload, with a link to the partner's data-use policy. We do not store your images and we do not train on them. You should always know exactly where your photo travels.
For people who would rather not upload at all, we built Door 1: a fully local, on-device experience that runs the model in your browser using WebGPU. Your photo never leaves your machine. It is the most private kind of AI portrait you can make — a true local AI avatar generator — and it is the clearest expression of our promise that your face stays your face. You can read how it works and try it yourself on Door 1 (local, private).
Our one promise: your face stays your face — no whitening, no warping, no training on your photo. The Ghibli-style trend showed how magical AI portraits can be. LumTale exists to make sure that magic is fair to everyone, not just the people the training data already favored.
The Ghibli-style AI portrait trend proved that people love seeing themselves reimagined — and that makes the bias in most tools all the more worth fixing. You should not have to choose between a beautiful portrait and a portrait that looks like you. With Lumling, you get both: a warm, distinctive style and a face that is unmistakably, faithfully yours.
If you care about fair, private AI portraits, we would love to have you along for what we build next. Join the waitlist for early access to new Lumling styles and early-access pricing — and if you want to start right now with zero upload, give Door 1 a try.
Join the LumTale waitlist for early access to new Lumling styles — and try Door 1 today for a private, on-device portrait that never leaves your phone or laptop.
Join the waitlistOr start private now: Try Door 1 (local, private)