Don'tjusttryiton.Prove it works.
Other try-ons ask “do you like how this looks?” We ask “does this look actually do what you need?” We analyse your skin, build a look for your specific concerns, then re-analyse the result to measure the change.
- Measured before/after
- Honest about limits
- Photos auto-deleted
Try-on tells you how it looks. Not whether it works.
A pretty filter can't tell you if the concealer actually hides your dark circles, or if that foundation just turned you two shades off. We close the loop with measurements.
- Answers: do I like it?
- No idea if your concern improved
- No shade-drift check
- Buy first, find out later
- Answers: does it do what I need?
- Per-concern before/after, measured
- Tone-drift and intensity penalties
- Honest about what makeup can't fix
One selfie. Six steps. A verdict you can check.
Every YouCam call is async: upload, create a task, poll. The orchestrator chains five of them into a closed loop and caps it at two rounds to protect your budget.
Selfie + quality gate
Face detection, blur, brightness and head pose run client-side. A bad photo is rejected before it costs a single credit.
client-side · MediaPipeBad photo? Rejected here, before any credit is spent.
Baseline analysis
Skin-analysis scores every concern, and tone analysis captures your skin, eye, lip and hair colours as the shade reference.
skin-analysis · skin-tone-analysisPlan candidate looks
A rule-based planner turns your top concerns into K distinct, layered looks, with shades derived from your measured tone.
planner · concern → strategy rulesRule table: dark circles → under-eye correction · redness → colour-correct · shine → T-zone powder
Render, then verify
Each look is rendered, then the same analysis runs on the render. Not a guess about the result: a measurement of it.
makeup-vto → skin-analysisScore, pick, refine once
Deterministic scoring weighs coverage against tone drift and over-application. Weak coverage triggers one refinement round, then it stops.
scoring engine · in code, never in the LLMTop-concern reduction under 20%? One refinement round with different patterns, then stop. Capped at two rounds.
Explain + shop
An explainer reads the numbers your code computed and turns them into plain language, plus shade-matched products to shop.
explainer · reads numbers, invents noneDon't compare one number. Weigh three.
Coverage, tone drift and intensity, with weights from what bothers you most. All deterministic and in code. Drag the controls and watch the winner change. This runs the real scoring function.
Punishes mismatched shades & mask-like finish
Stops the engine winning by over-covering
Test-retest noise: smaller changes count as zero
Sample numbers for illustration. In the app, every value comes from analysing your own render. Notice how “Full cover” reduces the most but loses once you care about shade match.
A try-on that only watches the number can't tell the difference.
We tested it on a real photo. A foundation a few shades too light scored about as well as a real concealer on dark circles, because it simply made the whole face lighter. Same number, very different result. So every run includes that shortcut as a control, and the shade check is what tells the two apart.
A good score, for the wrong reason. The under-eye wasn't treated; the whole face was brightened. A number-chasing try-on would happily recommend this.
Nearly the same score, and we can stand behind it: a real concealer treated the under-eye and the shade stayed close to your skin. We recommend this one, and say plainly that the effect is modest.
One real photo, one run, YouCam's raw skin-analysis score. The one-point gap between them is inside the measurement noise, and that is the point: the score can't separate them. In the app the control runs on your photo every time, and we only say it “fooled” a score when it beats our pick by more than the noise.
We'll tell you what makeup can't do.
Every concern is mapped to a strategy, or flagged as something makeup won't fix. Wrinkles and eyelid sagging are never scored, so we never pretend a filter solved them.
Dark circles
Under-eye correction + concealer.
Redness
Colour-correcting base on cheeks and nose.
Dark spots
Spot concealing + even base.
Acne & marks
Spot concealing + even base.
Shine
Setting powder across the T-zone.
Visible pores
Blurring base can soften the look. Makeup can blur pores optically, but it can't change them.
Uneven texture
Smoothing base. A smoothing base softens the look; heavy layers can emphasise texture.
Eye bags
Highlight + conceal to flatten shadow. Makeup can only reduce the shadow an eye bag casts, not the puffiness.
Wrinkles
Makeup tends to settle into lines. Skincare or a professional is the right lever.
Eyelid sagging
Eyelid structure isn't something makeup changes.
Predicted is a promise. Measured is proof.
Re-analysing a synthetic render measures the render, not your real skin. So there's a second stage: apply the look for real, take a selfie, and we run the same analysis and show predicted vs. actual.
Illustrative numbers. In the app, every value is computed from your own photos.
Apply it for real
Use the recommended look, take a fresh selfie in similar lighting.
Same analysis, again
The identical concern set runs on your real face, no simulation.
Calibration, over time
Aggregated across users, the predicted-vs-actual gap becomes a signal that sharpens the model. That's the roadmap.
Built to be honest with you.
Judges, and users, respect limits that are stated over biases that are hidden.
Consent first
Nothing is uploaded until you explicitly agree, and you can see exactly what happens to your photo.
Short-lived storage
Photos and renders auto-delete after a short window. Only hashed analysis results are ever cached.
One-tap delete
Delete my photos wipes your session and cached analyses immediately. No account needed.
Keys stay server-side
A thin proxy holds the YouCam key, so your browser never sees it.
Cosmetic, not medical
Results are cosmetic visibility estimates. Never a diagnosis, never dermatological advice.
Skin-tone caveats
If verification may be less reliable on deeper tones, we say so in your results instead of hiding it.
Stop guessing.
Start proving.
One selfie, about a minute, a measured before/after for the concerns you actually care about.