LEVEL / preparing
Loading LEVEL.
The questions people actually ask about this category, answered for LEVEL specifically — including the answers that are inconvenient. Every number here is read from the code that enforces it.
No. LEVEL never produces a rating, a ranking, a percentile, or any single number standing for how you look. It describes presentation choices — hair silhouette, facial-hair boundary, eyewear, near-face colour — and how confident it is about each observation. A score is the one thing this product will not produce, because a number attached to a face invites a comparison the evidence cannot support.
The reversible parts do, and they are the least discussed. Hair silhouette, facial-hair shape, eyewear, grooming and near-face colour change how a face reads, cost little, and undo if you dislike them. The claims that do not hold up are the ones promising a fixed ideal or structural change from a routine. LEVEL works only on the reversible half, and publishes the evidence behind each category in its methodology registry.
It depends entirely on what happens to the photograph, and most apps in this category do not say. The questions worth asking any of them: is the photograph stored, for how long, is it used to train a model, and does the privacy policy use the word biometric at all. LEVEL stores photographs in private object storage that is never publicly addressable, destroys them after twelve months without use, and does not use them to train any model.
Twelve months from your last capture or completed mission, after which they are destroyed automatically by a scheduled job — not on request, and not only if you remember to ask. The retention period is 365 days, it runs daily, and a record is kept that the deletion happened without keeping what was in the photographs. You can also delete everything immediately from your account.
Not at LEVEL. Photographs are sent to OpenAI for analysis and preview generation and to independent validators that check a generated preview is still recognisably you, and none of it trains a model. This is worth checking elsewhere too: at least one funded competitor's privacy policy states that facial data feeds machine learning algorithms.
No. Nothing surgical, nothing structural, nothing that changes skin tone, apparent age or identity. Every recommendation is reversible, which is the boundary the product is built around rather than a disclaimer attached to it. It is also not medical advice and does not diagnose anything.
18. Age is established from a date of birth recorded on the account rather than from a tick box, it is checked again at every point a photograph is accepted, and an account that has once stated an age below the minimum cannot restate it. This category is heavily used by teenagers and at least one of the largest apps in it ships at a 9+ rating.
It opens a country only once it can meet the rules that country sets for photographs of a face. Most of the world is open; Illinois needs a written release under its Biometric Information Privacy Act, the European Union and the United Kingdom each need a designated representative and an impact assessment, and Quebec needs a filing before a biometric database exists. You are asked where you live before any photograph is collected, and told exactly what is missing if your answer is somewhere not yet open.
Those produce a score; LEVEL refuses to. Beyond that: LEVEL is 18+ and enforces it from a recorded date of birth, publishes a twelve-month destruction schedule for photographs, names every processor that receives one, labels AI-generated images as generated rather than as an 'after', and declines to operate where it cannot meet the local rules. Those are verifiable claims rather than positioning — the methodology registry and the privacy policy say the same things in more detail.
How much the underlying evidence supports that specific observation, for your photographs. A low value is not a hedge, it is a refusal to assert: findings below the threshold are withheld rather than shown, and a released result states what was withheld and why. Most of this category presents every output with equal certainty, which is the part that makes it feel authoritative and makes it wrong.
The longer versions are in the methodology registry, the research library, and the privacy policy. A complete worked result, built from demonstration data, is at examples.
A confidence-aware presentation profile, five reversible directions and a twelve-week plan. No rating, no ranking, nothing surgical. Photographs are destroyed after twelve months and never train a model.