How to Upload Your Photo for Golden Ratio Face Analysis (And Actually Get Good Results)
Everything you're about to read exists because of one frustrated email.
Someone used our face analysis tool and got a 62%. They were disappointed. Then they tried again — same day, different photo — and got 74%. Same face. Twelve-point swing. They wrote to us asking which result was real.
Both were. That was the problem.
The score changed because the photo changed. Different lighting. Different angle. Slightly different expression. And those subtle differences shifted where the AI placed its landmarks, which shifted the measurements, which shifted the score.
Your golden ratio score isn't a truth about your face. It's a truth about your face in that specific photo. And that means the photo matters more than most people realize.
Here's how to get it right.
Why Photo Quality Makes Such a Huge Difference
When the AI analyzes your face, it places 468 tiny landmarks — dots on the corners of your eyes, the edges of your lips, the curve of your jaw. Everything gets measured from those dots.
Here's the thing: if a single dot is off by just 3 pixels, it can swing an individual ratio from 1.62 to 1.58. That's the difference between scoring 95% and 80% on one measurement. More on how the AI places these dots.
Three pixels. That's less than the width of a pencil tip on your screen.
So when people ask "why did my score change?" — it's not a bug. It's the tool being honest about how much a photo can distort what it sees.
The Setup That Gives You the Best Results
Camera Choice
Use the back camera of your phone. Not the front one. Not the selfie camera.
Why? Selfie cameras use wide-angle lenses. They stretch whatever's closest to the lens. Usually that's your nose. So in a selfie, your nose looks 10-15% wider than it actually is, while the sides of your face look narrower. That alone throws off at least two of the five measurements.
The back camera has a more neutral lens. It sees your face closer to how it actually looks.
Distance
Have someone else take the photo from about 3 to 5 feet away. Arm's reach plus another arm's length.
Too close and the lens distortion gets worse. Too far and your face is a small cluster of pixels — not enough detail for accurate landmark placement.
Height
Camera at eye level. Straight across.
Held too high? Your forehead looks massive and your chin disappears. Too low? Your jaw dominates and your forehead shrinks. Either way, you're measuring a distorted version of your face.
I learned this one by accident. First photo — taken from slightly above, standing — I got a 68%. Second photo — eye level, sitting — I got 73%. Same face. Just camera height.
Lighting (This One Matters More Than You Think)
Best option: Stand near a large window during the day. Face the window. The light should hit your face evenly from the front.
Second best: Go outside on an overcast day. Cloud cover acts like a giant softbox — diffused, even light from every direction. No harsh shadows.
What to avoid:
- Direct sunlight. Creates hard shadows under your nose, jaw, and brow ridge. Those shadows confuse the AI into placing landmarks at odd positions.
- Overhead fluorescents. Same shadow problem, especially under the eyebrows and nose.
- Backlight. If the light is behind you, your face is dark, and the AI has less contrast to work with.
Shadows are the #1 reason for inconsistent scores between photos. If you're getting wildly different numbers each time, check your lighting before blaming the tool.
Face Position and Expression
Expression: neutral
Mouth closed. Lips relaxed. Eyes open naturally. That's it.
A smile changes everything — it widens your mouth, narrows your eyes, pushes your cheeks upward. Three of the five measurements shift from a single smile. So even if you love how you look smiling, save it for a different photo.
Head position: straight
Face the camera dead-on. No tilt. No rotation. Even five degrees of head turn changes the apparent distance between your eye corners, which directly affects the eye spacing ratio.
And chin level — not tilted up, not tucked down. This one's harder than it sounds. Most people naturally tilt slightly without realizing it. Have the person taking the photo give you a heads up if you're off.
Hair: pull it back
The AI needs your hairline to measure face height. If your hair covers part of your forehead, the algorithm either guesses the hairline or defaults to the visible edge of your hair — which shortens your face height measurement and changes your main ratio.
Bobby pins, a headband, just push it back. Two seconds of effort, much better data.
Accessories: ditch them
Take off glasses. Frames near your eyes can be detected as facial features, shifting where the AI thinks your eye boundaries are. If you always wear glasses and want to see the difference, run the test with and without. You might be surprised.
Hats, headbands covering the forehead, big earrings hanging near the jawline — all of these can interfere.
Technical Stuff for the Photo Itself
Resolution: Minimum 640×480 pixels. Ideal is 1080×810 or higher. More pixels = more precision.
Format: JPG, PNG, or WebP all work. Avoid GIFs (they compress facial detail into oblivion).
Filters: No. Just no. Beauty mode, skin smoothing, face slimming — all of these change the contours that the AI uses for landmark placement. You'll get a score for the filtered version of your face, not your actual face. And what's the point of that?
Multi-face photos: Upload a photo with just your face visible. Group photos confuse the detection — it might map the wrong person's face, or try to average features across faces. Take the test with a proper setup for the most accurate results.
Testing Multiple Photos (Recommended)
Here's what I suggest for anyone who wants a reliable number:
Take 3 to 5 photos under slightly different conditions. Not dramatically different — don't switch from daylight to candlelight. Just minor variations:
- Same session, slightly different head positions
- Window light vs. outdoor light
- Taken by different people if possible
Run each photo through the calculator. Look at the scores. They should be within about 3-5 points of each other.
If you get one score that's 15+ points away from the others, that photo has a problem. Could be angle, lighting, or expression. Don't average it in — just toss it and use the consistent ones.
The average of your consistent scores is the closest thing to your "real" measurement. It smooths out the noise from individual photo conditions.
After the Results
Your results page shows five separate ratios, each with its own score. Please don't just look at the big number.
The individual breakdowns are way more interesting. You might find that your eye spacing is almost perfectly golden while your nose-to-mouth ratio is way off. Or that your facial thirds are balanced but your face shape is more rectangular than phi predicts.
Those details tell a story about your face's geometry. The overall number is just a summary — and summaries always lose nuance.
If you haven't tried it yet, go upload a photo and see what you get. Takes 30 seconds. And now you know how to take a photo that actually gives you something worth looking at.