Photo Conditions and Face Scores — What Changes the Reading

FaceRater reads geometry in one photograph. The result belongs to that image, not to a person as a fixed property. Lighting, framing, lens, camera distance, and head angle can change which landmarks are visible and whether a photo passes the quality gate.

This page records a small, controlled check using an existing repository demo image. It is a reproducibility note, not a clinical study, a population estimate, or evidence that a lighting change causes a particular score change for everyone.

What we tested

The fixture was the existing repository asset unsplash-1507003211169-0a1dd7228f2d.jpg. The file was 1000 × 1500 pixels before upload and has SHA-256 343afcdc5747f0de8c89b9cbbc40b78e6ff54d93be7bfc6fe70a8700b2946cf1. The image is a public Unsplash demo already present in the repository; no new person's photo was collected for this check.

The local transformations were deterministic and applied to copies of the same file:

• Brightness −35%: multiply pixel brightness by 0.65; geometry and dimensions stayed the same.
• Center crop 90%: remove 5% from each edge, then resize back to 1000 × 1500; this tests framing, not a different camera distance.
• In-plane rotation +5°: rotate the whole image around its center with a neutral fill; this tests a 2-D roll transformation, not a real 3-D head turn or a new pose.

The test ran through fresh Chrome contexts and the browser MediaPipe pipeline on 23 September 2026 (Los Angeles time). The free tools perform detection and measurements on the device.

Recorded observations

The original was uploaded twice on each route. The same baseline values were returned each time: one face, analysis size 853 × 1280, roll −1.945°, yaw proxy 0.0215, and blur variance 439.899.

• Original — no transform. The homepage returned overall reference score 94 with symmetry mean mirror deviation 0.837%, largest paired deviation 1.859%, symmetry score 90, proportions 100, and eye-area harmony 94. The golden-ratio tool returned 1.229, 2.563, and 1.114 for its three supported ratios.
• Brightness copy — −35% brightness. The homepage returned score 95; symmetry mean deviation 0.747%, largest pair 1.716%, yaw 0.01995, and blur variance 188.002. The golden-ratio values were 1.226, 2.577, and 1.108.
• Center-crop copy — 90% crop, resized. The homepage returned score 94; symmetry mean deviation 0.933%, largest pair 1.878%, yaw 0.01835, and blur variance 459.909. The golden-ratio values were 1.232, 2.530, and 1.135.
• In-plane rotation copy — +5° rotation. The homepage returned score 94; symmetry mean deviation 0.858%, largest pair 1.549%, roll −6.777°, yaw 0.02354, and blur variance 487.244. The golden-ratio values were 1.250, 2.560, and 1.110.

These are repeated observations from one public demo image. They show that the same face geometry can produce slightly different photo-space values after a pixel transformation, while the tool still accepts this frontal fixture. They do not establish a general brightness, crop, or pose effect, and they are not calibration data.

What this does and does not test

The baseline's roll and yaw values are measurements of the uploaded photograph. The 2-D rotation copy is a pixel transformation only. We did not capture the same subject at a new camera distance, with a different lens, under a newly arranged light, or with a true 3-D head turn. We therefore do not claim that this page measures those real-world changes.

For a more useful retake, face the camera, keep the whole face visible, use even light, and avoid a close wide-angle selfie. If the tool rejects the image, retake it rather than treating a forced score as information.

How to read the tools

The face symmetry test fits a facial midline and compares six actual left/right landmark pairs after in-plane roll correction. The golden ratio face calculator draws three supported 2-D ratios and shows their formulas. The value 1.618 is a reference constant for those ratios, not a universal facial target or a measure of attractiveness.

See the methodology for formulas and reference-band limits, and the privacy policy for the difference between free local analysis and the separate consented paid visual flow.

Continue with your result

Read how to interpret a face rating, compare AI analysis with human feedback, or learn how to use the styling report. Each answers a different next-step question.