I run an app called KOOTD — an AI fashion app that rates your outfits. This time I added a new feature to it that looks at your face: Glow-Up AI. I won't pretend it's the right answer. I don't actually use face-rating apps myself — I researched a bunch of them to build this, and looking through them, a few things made me want to do it differently.
A few things bugged me while researching them
First, the scores are too generous. Everyone gets a 7 or an 8. It feels nice for a second, but it's hard to tell what the number actually means.
Second, there's no reasoning. "Cheekbones: 79." So what's the user supposed to do with that?
And a lot of them rate the bone structure you were born with — jaw, cheekbones, symmetry. But you can't change that. A low number just stings, and a high one leaves you nowhere to go.
I'm not saying they're bad apps. They just weren't the direction I wanted to go, so I tried building something different.
First, I tried to be honest (not objective)
Honestly, AI beauty scores are biased by nature, and there's no "correct answer" here to begin with. So KOOTD doesn't claim to be objective truth.
What I tried for was an honest read instead of a flattering number. Average should look like average. Most people land somewhere in the middle, and that's normal — so showing it that way felt like the right thing to do. I tried to put the kindness in a line of feedback, not in an inflated score.
I did want to show why
A bare number with nothing behind it was the most frustrating part for me. So next to each dimension I added an observation — one line pointing to where the number came from.
"Overall smooth and even in tone, with slight texture visible under lighting."
At least that way it makes sense. And the score stops being about some abstract face and starts being about you.
Six dimensions instead of sixteen
The existing apps throw a lot of categories at you. Sixteen of them, overlapping, and even looking closely I couldn't always tell what they meant.
So I spent a while looking through other face-rating apps and some aesthetics and psychology material, and kept only six that don't overlap — only what a single front-facing photo can honestly judge.
| Dimension | What it looks at |
|---|---|
| Skin | Clarity, evenness, blemishes, shine — skin texture |
| Hair | Cut, health, styling, volume (within your natural hair) |
| Grooming | Brows, stray hairs, facial hair — the "tidiness" layer |
| Contour | The visible face line, head-on |
| Harmony | Proportion, symmetry, overall balance |
| Features | The eyes, nose, and lips themselves |
The first four (Skin, Hair, Grooming, Contour) are things you can actually change; the last two (Harmony, Features) are your baseline. And whatever a front photo can't measure accurately — like the depth you'd need a side profile for — I didn't force a score on. Pretending to see what you can't is the worst thing you can do here.
"Glow-up" means only what you can change
This is the part I cared about most while building it. I didn't want to stop at the score.
- Upload one photo and you get the six scores and observations.
- Polish the changeable axes and it shows how far you could go, as a potential score.
- It doesn't leave that as words — it renders a better version on your actual photo. You can drag a before/after slider and save it.
- A haircut it flags as suiting you can be tried right inside the app.
I wanted it to be a mirror that shows your next step, rather than a scorecard that knocks you down.
Trying not to dock anyone's face
This is one thing I really wanted to hold to. It doesn't deduct points for how close you are to any ethnicity or Western standard. Monolids, broad noses, full lips — none of those are treated as flaws. Age, lines, grey hair, scars, marks: not deductions, just part of you. I tried to look at health and grooming within your face, not your distance from some standard.
A little fun, too
Numbers alone are kind of dry, so I added a vibe / animal-face read, your personal color, and your face shape. The result wraps into a 9:16 card you can share with a friend. Post it to the community and people see the per-dimension scores and observations too — so it's a result you look at together, not one you close alone.
There's still plenty that's rough. But if you give it a try and something feels off or lacking, I'd really appreciate you telling me. I'll keep fixing it, slowly.
You can scan your face over at KOOTD.