Every source Forma uses, with what was studied and what it cannot prove about your photo.
14 original papers behind the photo guidance, measurements and suggestions. Each title opens the publication.
Browse all sources and guidanceA study of 96 adults compared 2D photo measurements with direct measurements. Agreement differed by dimension; cheek and jaw widths showed meaningful differences.
Studied: 96 adults recruited in two districts in Malaysia
A review of 27 studies examined optical facial measurement systems. Protocol, movement and landmark placement affect results.
Studied: 27 studies of 3D optical systems measuring living people or inanimate face-like objects
A twin study found that individual differences in face preferences were largely associated with individual environments. Shared preferences also existed. Geometry alone cannot produce a universal attractiveness score.
Studied: 547 identical-twin pairs, 214 same-sex nonidentical-twin pairs, and 660 other Australian adults
Using published average facial dimensions, the authors modeled how a 30 cm selfie enlarges apparent nasal breadth relative to a portrait from 1.5 m. Forma recommends a consistent, more distant camera setup.
Studied: Geometric model using published average facial dimensions; no participant photos were tested
In a randomized Australian trial, daily sunscreen use reduced progression of measured skin aging compared with discretionary use over 4.5 years.
Studied: 903 Australian adults younger than 55 years
Observers rated photographs after restricted sleep differently from photographs after normal sleep. Keeping sleep and capture conditions consistent makes comparisons more interpretable.
Studied: 25 photographed adults and 122 independent raters in Sweden
In 155 selected Iranian medical students, Table 4 reports internal G–Sn–Pg angle mean 167.674° and SD 4.1937°. Forma shows this geometric reference alongside your profile measurement.
Studied: 155 Iranian medical students selected for Class I dental occlusion and no prior orthodontic treatment
Men were photographed clean-shaven, after 5 and 10 days of growth, and after at least six weeks for a full beard. On average, the women surveyed rated heavy stubble highest and light stubble lowest.
Studied: Ten photographed men of European descent; ratings from 177 men and 351 women in the main comparison
After a short night followed by 31 hours awake, photos were rated slightly less attractive (38 vs 40 on a 100-point scale), less healthy and more tired than after normal sleep.
Studied: 23 photographed adults and 65 independent raters
Changing apparent camera distance altered social judgments of the same face, including attractiveness ratings.
Studied: 18 unfamiliar White male faces in the initial photo set, judged by separate groups of adult viewers; later experiments used new photos or online viewers
Thirty volunteers were photographed at close selfie distances and at a longer clinical distance. Some facial dimensions differed between setups.
Studied: 30 volunteers photographed in three camera setups
The same 45 people were photographed at three focal lengths; facial shape measures and observer judgments changed across the images.
Studied: 45 photographed adults and 369 student raters
Posing with an upward or downward tilt changed apparent facial width-to-height ratio and how intimidating the pictured person seemed to raters. It informs Forma's caution about head position.
Studied: 57 White adults photographed in Study 1; 101 White adults rated selected target photos in Study 2
Across photos of the same people, expression and camera parameters changed measured facial width-to-height ratio.
Studied: Five White male actors in film stills, 21 photographed men in a distance study, and 20 men from a facial-expression image set
Camera distance, light, pose and expression shift what a picture shows. Forma checks the setup and asks you to compare like with like.
Reference samples describe the people and methods in that study. They are not a universal ideal or a personal verdict.
It predicts how raters might see a front photo. On 102 test faces it agreed with them at r = 0.58, less closely for men's faces, so it never counts as progress.
All 27 sources: original articles, professional guidance, dataset terms, methods and technical documentation.
Why Forma links it: Repeatability and accuracy are different questions.
A study of 96 adults compared 2D photo measurements with direct measurements. Agreement differed by dimension; cheek and jaw widths showed meaningful differences.
This does not validate Forma's model-assisted measurements. Repeated values show their observed spread, not clinical accuracy or a guaranteed physical change.
Research articleRead at doi.org
Why Forma links it: Evidence for 3D devices does not validate a 2D selfie.
A review of 27 studies examined optical facial measurement systems. Protocol, movement and landmark placement affect results.
These are different capture systems. Their accuracy figures are not error bounds for this app.
Research articleRead at pubmed.ncbi.nlm.nih.gov
Why Forma links it: A close-up selfie can change apparent proportions.
Using published average facial dimensions, the authors modeled how a 30 cm selfie enlarges apparent nasal breadth relative to a portrait from 1.5 m. Forma recommends a consistent, more distant camera setup.
This was a geometric model, not a test of participant photos or Forma's measurements. It does not quantify distortion for a particular phone or face.
Research articleRead at pmc.ncbi.nlm.nih.gov
Why Forma links it: A profile angle can be compared with a named sample.
In 155 selected Iranian medical students, Table 4 reports internal G–Sn–Pg angle mean 167.674° and SD 4.1937°. Forma shows this geometric reference alongside your profile measurement.
Selected Class I occlusion sample, standardized photography; not a universal ideal or beauty rating. No individual data or median were supplied, so no empirical percentile is calculated.
Research articleRead at pmc.ncbi.nlm.nih.gov
Why Forma links it: Passport-photo standards set the capture limits Forma checks.
For reference face images, ICAO recommends a camera 1.0–2.5 m away at eye level, head pitch and yaw within ±5°, roll within ±8°, and even light across both cheeks. It notes that selfie-style portraits are unlikely to meet the minimum distance.
Written for face recognition, not facial measurement. Forma uses these limits as guidance, and its automatic angle estimates are approximate.
Guidance or technical sourceRead at icao.int
Why Forma links it: The face model's published error depends on capture mode.
On 1,700 front-camera selfies, mean landmark error was 3.24% of interocular distance on the first frame (reacquisition) and 2.62% while tracking. The card separately reports 2.56% disagreement among human annotators on 58 images and warns against faces rolled more than 8° from upright.
These are averages across landmarks and images, not a precision estimate for Forma's still-photo workflow or repeat photos of one person. The 8° warning concerns roll, not a validated yaw or pitch limit.
Guidance or technical sourceRead at storage.googleapis.com
Why Forma links it: One forensic protocol estimated 5% uncertainty.
In forensic photo comparisons, landmark placement and photography contributed to an estimated 5% overall uncertainty for selected facial proportions. Forma uses that number only as a provisional minimum for front-ratio repeat-photo bands.
This was not a study of Forma or phone check-ins. The thesis concluded that anthropometric image comparison was not adequate as courtroom identification evidence; Forma still needs its own repeat-photo study.
Doctoral thesisRead at theses.gla.ac.uk
Why Forma links it: Perspective distortion from camera distance affects face judgments.
Changing apparent camera distance altered social judgments of the same face, including attractiveness ratings.
A controlled perception study, not a correction formula for one person's photo.
Research articleRead at journals.plos.org
Why Forma links it: Close-up selfies can alter apparent dimensions.
Thirty volunteers were photographed at close selfie distances and at a longer clinical distance. Some facial dimensions differed between setups.
Camera distance and camera type both changed between selfie and clinical photos, so this study cannot isolate distance alone or calibrate Forma measurements.
Research articleRead at pubmed.ncbi.nlm.nih.gov
Why Forma links it: Camera setup can change apparent facial shape.
The same 45 people were photographed at three focal lengths; facial shape measures and observer judgments changed across the images.
The camera settings were controlled. This does not quantify error in Forma's phone-photo workflow.
Research articleRead at journals.plos.org
Why Forma links it: Head tilt changes apparent facial width-to-height ratio.
Posing with an upward or downward tilt changed apparent facial width-to-height ratio and how intimidating the pictured person seemed to raters. It informs Forma's caution about head position.
This concerns one ratio and an impression judgment, not a threshold for personal progress or a change in underlying facial structure.
Research articleRead at doi.orgOpen free full text
Why Forma links it: The same person's measured proportions vary across photos.
Across photos of the same people, expression and camera parameters changed measured facial width-to-height ratio.
The studies used images of White men and measured one facial ratio. Their effect sizes do not transfer to every Forma measurement or camera setup.
Research articleRead at pmc.ncbi.nlm.nih.gov
Why Forma links it: Attractiveness has shared and individual components.
A twin study found that individual differences in face preferences were largely associated with individual environments. Shared preferences also existed. Geometry alone cannot produce a universal attractiveness score.
This does not mean that a fixed percentage of every beauty judgment is subjective.
Research articleRead at pmc.ncbi.nlm.nih.gov
Why Forma links it: Beard stages take days to weeks to reach.
Men were photographed clean-shaven, after 5 and 10 days of growth, and after at least six weeks for a full beard. On average, the women surveyed rated heavy stubble highest and light stubble lowest.
Average preferences of one rater sample for one set of models. Your own preference and context can differ.
Research articleRead at doi.org
Why Forma links it: Consistent sun protection supports long-term skin quality.
In a randomized Australian trial, daily sunscreen use reduced progression of measured skin aging compared with discretionary use over 4.5 years.
The endpoint was skin aging, not attractiveness, and results are not a guaranteed score improvement.
Research articleRead at pubmed.ncbi.nlm.nih.gov
Why Forma links it: Choose broad-spectrum SPF 30 or higher.
Use broad-spectrum, water-resistant sunscreen along with shade and protective clothing. Follow product directions and reapply during outdoor exposure.
Individual skin conditions and sensitivities may need personalized care.
Guidance or technical sourceRead at aad.org
Why Forma links it: Keep cleansing gentle and avoid scrubbing.
Use a gentle cleanser and lukewarm water, apply with fingertips, and moisturize if skin is dry or itchy. Irritating routines are not a shortcut to better skin.
Photographs cannot diagnose acne, rosacea or other skin conditions.
Guidance or technical sourceRead at aad.org
Why Forma links it: Acne treatments take weeks to show.
If a treatment works, you should notice some improvement in 4 to 6 weeks, and clearing can take two to three months or longer. Switching products every few days can irritate skin.
Forma doesn't diagnose skin conditions or recommend treatments. Painful, scarring or persistent acne deserves a dermatologist.
Guidance or technical sourceRead at aad.org
Why Forma links it: Sleep restriction can affect how a face is perceived.
Observers rated photographs after restricted sleep differently from photographs after normal sleep. Keeping sleep and capture conditions consistent makes comparisons more interpretable.
A small, short-term experiment. It does not predict a personal change in attractiveness.
Research articleRead at pmc.ncbi.nlm.nih.gov
Why Forma links it: Hair grows about a centimetre a month.
Scalp hair grows about 1 cm a month on average, ranging from 0.6 to 3.36 cm. Beard hair grows about 0.38 mm a day and eyebrows about 0.16 mm a day.
Average growth rates. They don't say when a haircut or beard looks best.
Guidance or technical sourceRead at atsdr.cdc.gov
Why Forma links it: Sleep loss changes ratings only a little.
After a short night followed by 31 hours awake, photos were rated slightly less attractive (38 vs 40 on a 100-point scale), less healthy and more tired than after normal sleep.
23 people and acute sleep deprivation. It doesn't show that extra sleep improves appearance in people who already sleep normally.
Research articleRead at pmc.ncbi.nlm.nih.gov
Why Forma links it: Mewing is not an established way to reshape an adult jaw.
The AAO cautions that claims of jaw reshaping through mewing lack scientific support. Concerns about bite, jaw function or dental alignment belong with an orthodontist.
No jaw-training, hormone, surgery or weight-loss prescription is generated from these photos.
Guidance or technical sourceRead at aaoinfo.org
Why Forma links it: The AI estimate comes from an openly licensed image model, not a beauty dataset.
SigLIP 2 matches images with text. Forma compares your front photo with fixed descriptions of attractive and unattractive faces, blends that with three measurements, and runs it all in your browser.
The model was not trained on attractiveness ratings, and its web-scale training data can carry social biases. It predicts how raters might see a photo, not beauty or worth.
Model repositoryRead at huggingface.co
Why Forma links it: The estimate's accuracy was measured on faces rated by thousands of people.
102 adults photographed in London, each face rated for attractiveness by 2,513 people. Forma's estimate agreed with the average rating at r = 0.58 (95% CI 0.44–0.70), close to a typical single rater (median 0.62).
Only 102 faces, about two-thirds White and mostly in their twenties (ages 18–54), so the figure is approximate. Agreement was lower for men's faces (r = 0.43). The set was used only to measure accuracy and scale the estimate; none of its photos ship with Forma.
Dataset repositoryRead at doi.org
Why Forma links it: Change one thing and revisit it over time.
A guide to planning and interpreting repeated within-person comparisons. It motivates recording a clear change and checking it over time.
Forma's grooming experiments are informal personal records, not clinical N-of-1 trials or proof of causation.
Guidance or technical sourceRead at effectivehealthcare.ahrq.gov
Why Forma links it: Automatic landmarks are starting points, not verified anatomy.
Documents the face landmarks and browser inference used to initialize Forma's editable photo measurements.
Landmark detection is not clinical anatomical annotation or an attractiveness assessment.
Guidance or technical sourceRead at developers.google.com
Why Forma links it: A reference mesh helps place approximate points.
The reference geometry used to check approximate correspondence between named facial points and MediaPipe vertices.
A mesh vertex is an estimate on an image, not a verified anatomical landmark.
Guidance or technical sourceRead at github.com
Formulas, landmarks and source notes sit next to the results they inform.
Open research in the workspace