Face analysis, on your phone
Join 5,000+ people who measured instead of guessed.
Measured.
Not guessed.
478 landmarks, 19 readings in degrees and ratios, and a 3D model of your face on Face ID iPhones. Your photo never leaves the phone.
Measured
478 landmarks, not an opinion
On device
your photo never uploads
Free first scan
readings before any paywall
478landmarks
placed on every face, in the browser or the app
19readings
angles and ratios, each scored against its band
3Don Face ID iPhones
a TrueDepth model you can turn and relight
Studies show
Most of what you see in the mirror
is below the threshold of noticing.
These are the published numbers Symmi scores against. Every reading is placed on a band drawn from measurements like these, not from an opinion.
6mm
A chin has to sit 6 mm off centre before an ordinary person notices. Eyelid 2 mm, mouth corner 3 mm, brow 3.5 mm, nose tip 4 mm.(1)
Wang, 2017. Aesthet Surg J.
93%
Of 100 people measured before eyelid surgery, 93% had at least 1 mm of left-right difference in brow or eyelid position.(2)
Macdonald, 2014. J Otolaryngol Head Neck Surg.
13→51%
The same men were judged attractive 13.5% of the time on a 50 mm lens and 51.1% on a 105 mm lens. Same faces, same light.(3)
Třebický, 2016. PLoS ONE.
30%
A selfie taken at 12 inches makes the base of the nose look about 30% wider than a photo taken from further back.(4)
Ward, 2018. JAMA Facial Plast Surg.
73%
73% of healthy newborns already had at least one head or neck asymmetry on the day they were photographed.(5)
Stellwagen, 2008. Arch Dis Child.
9.3mm
Jaw width read from a flat photo missed the caliper by 9.3 mm on average, with an agreement score of only 0.16.(6)
Lim, 2022. Front Public Health.
Evidence review
Does it actually work?
The trials, the reviews, the honest verdicts.
Every method people try on their face, graded on what has been measured. Most habits land on works a little. Styling changes the look, not the face. A few things only a clinic can do.
Chewing on both sides
for an even face
Works a little
A lifetime of one-sided chewing leaves the masseter about 1 mm thicker. Muscle only; the 748-person cohort found no effect on facial asymmetry.
Rajesh, 2024. Cureus; NFBC1986, 2024.
Read the review →
Chewing gum
for a sharper jaw
No evidence
58 adults, three sessions a day for six months. Bite force rose; masseter thickness and jaw shape did not change.
Jung, 2024. J Oral Rehabil.
Read the review →
Mouth taping
for face shape
No evidence
Ten studies, 213 patients in total, every one rated poor, none measuring a face. The monkey study used growing animals.
Rhee, 2025. PLOS ONE.
Read the review →
Mewing
for the jaw
No evidence
Zero trials. Real appliances worn daily by 1,251 growing children moved the jaw relationship by 0.02 degrees.
Batista, 2018. Cochrane.
Read the review →
Same face, different habits
What changes
when the habits change.
Drag the line. Left is the untreated version: one-sided chewing bulking one masseter, a heavier eyelid, a dropped mouth corner, fluid under the chin, a tilted head. Right is the same person with the muscle evened out, the swelling gone, the head level and the grooming done. Bone is identical on both sides.
- [1]Masseter · chew on both sides, the bulk evens out
- [2]Eyelid and brow · sleep, salt and screen time show here first
- [3]Mouth corner · posture and jaw rest position
- [4]Under the chin · fluid and body fat, the fastest layer to move
- [5]Head position · ears over shoulders, chin level
Example faces, generated for this page. An illustration of the layers that move, not a result or a promise.
BeforeAfter
BeforeAfterWhat you get back
Your face,
with the lines on it.
This is the photo after the free symmetry tool has placed its landmarks and drawn where each reading was measured. Real output on an example face; every number you get back points at one of these lines.
- [1]Midline · glabella to chin, corrected for head roll
- [2]Eye line · inner corners, tilt read against it
- [3]Brow line · height difference in % of face width
- [4]Mouth corners · level or dropped, and by how much
- [5]Chin offset · distance from the midline
How it works
A number you can check,
instead of an opinion you cannot.
The old way
- 01Ask a chatbot for a rating
- 02Stare at a mirror
- 03Ask a friend who says you look fine
- 04Fixate on one feature
- 05No baseline, so nothing to compare
The new way
- Step 1ScanOne front photo, framed the same way each time.
- Step 219 readingsAngles and ratios, each with the band it is scored on.
- Step 33D on Face IDA TrueDepth model you can turn and relight.
- Step 4RescanThe same readings again, so change shows up as a number.
- Step 5RoutineGua sha, ice and a 30-day plan for the parts that move.
Why photos disagree
Your camera
has an opinion.
Lens, distance and a few degrees of head turn change the face in the frame before anyone measures anything. Symmi fixes the framing first, then reads the numbers.
- LensThe same 45 faces were rated by 369 people at 50, 85 and 105 mm. Judged-attractive rates roughly tripled between the two ends, with nothing but the focal length changed.(3)
- DistanceHeld at 12 inches, a phone makes the base of the nose look about 30% wider than a photo taken from further back.(4)
- Head turnTurning the head 5° to 10° already distorts the size readings on a flat photo; roll and turn together also distort shape readings such as the jaw angle.(7)
- MirrorPeople reliably prefer the mirrored version of their own face, and their friends reliably prefer the unmirrored one. Neither of you is seeing the same face.(8)
What we measure
Nineteen readings,
five regions.
Free in your browser
Start with one reading.
Same face, three views
One face.
Three angles, three sets of readings.
The browser tools read the front view. The app adds the three-quarter view, and on Face ID iPhones builds the full 3D model. Example face, generated for this page.
Measured in the browser
Measured in the app
3D on Face ID iPhones
Precision
Retake the photo.
Same number.
Three example faces were each measured five times: the original, tilted ±1.5°, downscaled to 90% and darkened by 15%. The spread is half the range between the highest and lowest run, taken from the worst of the three faces.(10)
Full repeatability table →| Reading | Unit | Runs | Spread |
|---|---|---|---|
| Symmetry index | 0–100 | 3 × 5 | ±0.5 |
| Chin offset | % of face width | 3 × 5 | ±0.2 |
| Canthal tilt | degrees | 3 × 5 | ±0.4 |
| Jaw angle | degrees | 3 × 5 | ±0.4 |
| Jaw : face width | ratio | 3 × 5 | ±0.01 |
| fWHR | ratio | 3 × 5 | ±0.01 |
| Midface ratio | ratio | 3 × 5 | ±0.01 |
Measured 2026-09-12 with the exact code that runs on this site. Lens distance is not part of this test and moves readings more than anything here; take the same kind of photo each time.
Inside the app
Nineteen readings.
One model. One plan.
Every reading with its band
Angles and ratios, side by side
Gua sha, ice and a 30-day plan
Track it
Rescan next month.
Compare the numbers, not the feeling.
Every scan is stored on your phone with its readings. Rescan under the same light and framing and the app lays the two sets side by side, so a change shows up as a number rather than an impression. The routine sits next to it: gua sha, ice and a 30-day plan you tick off day by day.
- Scan historykept on device
- Comparisonreading by reading
- Routine30 days
Bone is given, habits are yours
Same DNA.
Different faces.
147 pairs of identical twins, eight facial measurements. The pairs were pulled apart by what they did, not what they inherited. Symmi cannot move bone. It measures the parts that habits reach, and gives you a routine for them.(9)
- Sleeping face downlinked to nose midline and mouth-corner asymmetry
- Tooth extractionlinked to a tilted bite plane
- Wearing dentureslinked to nose midline and mouth-corner asymmetry
- Smokinglinked to a tilted bite plane and a drooping upper eyelid
Reading
What actually shapes a face.
Privacy
Your photo
stays yours.
The browser tools place the landmarks and compute every reading inside your browser tab. Close the tab and the photo is gone. The app does the same on the phone: scans, readings and the 3D model are stored on the device, and nothing is uploaded to be analysed.
0
photos uploaded
On device
landmarks and readings
On device
scan history
FAQ
Straight answers.
One reading here.
Nineteen in the app.
References
- (1)Wang TT et al. Aesthet Surg J 2017;37(4):375-385.
- (2)Macdonald KI et al. J Otolaryngol Head Neck Surg 2014;43:36.
- (3)Třebický V et al. PLoS ONE 2016;11(2):e0149313.
- (4)Ward B et al. JAMA Facial Plast Surg 2018;20(4):333-335.
- (5)Stellwagen L et al. Arch Dis Child 2008;93(10):827-831.
- (6)Lim YC et al. Front Public Health 2022;9:813058.
- (7)Gateño J et al. J Oral Maxillofac Surg 2011;69(3):655-662.
- (8)Mita TH, Dermer M, Knight J. J Pers Soc Psychol 1977;35(8):597-601.
- (9)Liu MT, Iglesias RA, … Guyuron B. Plast Reconstr Surg 2014 Oct.
- (10)Symmi repeatability test, 12 September 2026. Three example faces, five runs each, measured with the code that runs on this site. Full table on the methodology page.