The short answer: Omoggle uses computer vision and facial landmark detection, not a large language model. The developer has made several public clarifications about the underlying technology since the platform went viral in May 2026.
What Omoggle Actually Uses
Omoggle uses two distinct AI systems for two different purposes:
1. MediaPipe for facial analysis
The scoring system uses MediaPipe FaceMesh, Google's open-source computer vision model, to map facial landmarks in real time from the webcam feed. FaceMesh produces roughly 478 3D landmarks per face in about 25 milliseconds, and it all runs client-side — in your browser, on your device. Those landmark coordinates become distances and angles, and those distances and angles become scores for symmetry, canthal tilt, jawline definition, cheekbone prominence, skin clarity, and overall harmony.
This is not a "thinking" AI. It is a measurement system that maps specific landmark coordinates and calculates ratios against reference values. The scores are mathematical outputs, not subjective judgments.
How the 5 sub-scores are weighted
The individual metrics don't count equally. Each sub-score is normalized to a 0–100 range, then blended with unequal weights before being mapped onto the 1–10 PSL band:
- Harmony and symmetry carry the most weight — they're stable traits that don't swing between sessions.
- Jawline and cheekbone structure sit in the middle (and are the most lighting-sensitive to capture).
- Skin clarity and canthal tilt are down-weighted, because skin especially can be distorted by lighting or a single bad night's sleep.
The final mapping to a 1–10 PSL score is non-linear: because most faces cluster in the middle of the distribution, the raw-score gap between PSL 7 and 8 is far larger than between 5 and 6. For the full breakdown, see the PSL scale chart.
2. NSFW.js for content moderation
Separately, Omoggle uses NSFW.js, an open-source machine learning model, to evaluate whether nudity is present in players' video feeds. This is the moderation layer — it detects prohibited content and can end or flag a match if triggered.
Developer Clarifications (May 2026)
Developer Pablo Rogers made two rounds of public clarifications following the platform's viral surge:
On data privacy: Facial landmark processing runs entirely on the user's device via MediaPipe. The raw video feed and facial mesh data are not transmitted to Omoggle's servers. Rogers stated: "Our intent is not to collect and sell user data, or to maintain broad archives of private user sessions."
On advertising: Despite the above, Omoggle's privacy policy reserves the right to use "match-related content, including clips, screenshots, gameplay visuals, audio, usernames, display names, profile images, rankings, and match results" for advertising. Rogers has not fully reconciled this with the data privacy statements.
Why Scores Are Inconsistent
Because Omoggle uses geometric measurement rather than holistic AI judgment, small changes in input conditions produce large changes in output scores. Camera angle, distance, lighting direction, and even webcam quality all affect how FaceMesh maps facial landmarks — and therefore what scores it generates. This is why the same person can score 5.2 in one session and 6.8 in the next. We break down all seven causes in why your Omoggle score keeps changing.
This also explains why camera angle optimization is the single most effective way to improve your Omoggle score. You are not changing your face — you are optimizing the geometric inputs to a landmark detection system. See: How to Win on Omoggle.
Is It Actually Fair?
No, not in any meaningful sense. Omoggle measures how well your face is captured by a specific computer vision system under specific conditions — not how attractive you are, not what your PSL score "really" is, and not anything that translates reliably to real-world perception. It is entertainment that uses the vocabulary of looksmaxxing, not a validated measurement of anything.
Test the same AI system offline — use our free AI Face Analyzer, which uses the same 6-metric scoring approach as Omoggle without requiring a live opponent.