How it works
Why the mirror lies
Lighting, sleep, mood and what you paid all change what you see. A ₹2,500 serum four weeks in feels like it should be working. We replace that with measurements and an honest rule.
Split-face: the strong design
You apply the product to one cheek only. Every photo then contains its own control: the other side of the same face, under the same light, on the same day. We find the face midline with MediaPipe Face Mesh and measure YouCam’s per-concern detection mask on each side. What we track is the gap between the sides compared with the gap at baseline. Published trials use this design for the same reason: Bissett et al. (2004) tested niacinamide on one half of the face against a control on the other, and Kafi et al. (2007) tested retinol on one arm against vehicle on the other. Split-face is used concern by concern: if your own repeat shots show a concern's left/right gap is more than twice as noisy as its whole-face score, that concern is judged on the whole-face trend instead, and its card says so.
Does it actually detect a change?
We tested the measurement with real YouCam calls on a licence-free stock face, through a fake camera. Across 6 untreated photos, the redness gap between the cheeks varied by only ±1.4 points. Then we added a known soft red tint to one cheek: the gap moved by 17.2 points, on the right side, about 22 times its standard error. Pores didn't move (correctly), but their gap jumped by ±13 between untreated photos, so pores fall back to whole-face scoring. This tests the measurement, not a skincare product, and it's one face, not a clinical study.
Your own noise
Two selfies taken a minute apart don’t score the same. Your baseline is 3 photos, and every 3rd check-in takes 2, so we can measure how much your scores wobble. Until there are enough repeats we use a starting guess of ±3 points and blend in your own as they arrive.
The rule
A concern is “Working” only when the change since baseline is bigger than 2× its standard error and bigger than 2× your noise, after at least 4 check-ins, and not before the product type’s earliest-effect day. “Getting worse” is the mirror image, and can show earlier, because stopping early is the safe side. Otherwise it’s “No effect beyond noise”, “Too early to tell” or “Not enough data”. The rule is plain code with tests. No AI decides your verdict.
Raw score, not UI score
YouCam returns two scores per concern. The “UI score” is reshaped to look friendlier: in our tests a raw 46 showed as 66, which squeezes real differences together. Every number here is the raw score.
What YouCam does
Each photo goes to the YouCam Skin Analysis API (HD): upload the file, create a task with the concerns you chose plus redness and texture as controls, then fetch the result. That costs 12 units per photo for up to 4 concerns. The result holds a raw score and a detection mask per concern.
Time to effect
- Moisturiser / barrier cream: judged from day 14; trials ran ~28 days. Hydration changes show within days; barrier-repair trials typically report at 2–4 weeks. (not yet checked)
- Niacinamide serum: judged from day 28; trials ran ~84 days. Bissett et al., Int J Cosmet Sci 2004: 12-week split-face trial, 5% niacinamide vs control, 50 women. Hakozaki et al., Br J Dermatol 2002;147:20–31: paired 5% niacinamide vs vehicle. ✓ checked
- Vitamin C serum: judged from day 56; trials ran ~90 days. Humbert et al., Exp Dermatol 2003;12:237–44: 5% vitamin C vs vehicle, assessed at 3 and 6 months. ✓ checked
- Retinol / retinal (OTC retinoid): judged from day 56; trials ran ~168 days. Kafi et al., Arch Dermatol 2007;143:606–12: 0.4% retinol vs vehicle on opposite arms, 24 weeks. ✓ checked
- Acne treatment (BPO, salicylic, adapalene): judged from day 28; trials ran ~84 days. Topical acne trials typically use 12-week endpoints (e.g. adapalene / benzoyl peroxide vs vehicle). (not yet checked)
- Exfoliating acid (AHA / BHA): judged from day 28; trials ran ~56 days. Glycolic / salicylic acid cosmetic trials typically report at 4–12 weeks. (not yet checked)
- Dark-spot treatment (azelaic, tranexamic, arbutin): judged from day 56; trials ran ~84 days. Hyperpigmentation trials (azelaic acid, tranexamic acid) mostly report at 8–12 weeks. (not yet checked)
- Redness / soothing (azelaic, centella): judged from day 28; trials ran ~56 days. Azelaic acid 15% gel rosacea trials (Thiboutot et al.) ran 12–15 weeks, with redness still improving at the end. (not yet checked)
- Oil control: judged from day 14; trials ran ~28 days. Sebum-control cosmetic studies commonly report at 2–4 weeks. (not yet checked)
- Something else: judged from day 28; trials ran ~56 days. Conservative default: one skin-turnover cycle (~4 weeks) before judging. (not yet checked)
Trial lengths come from the papers. The “judged from” day is our own conservative cut-off, set earlier than the trial endpoints.
Privacy
Trials, scores and photo thumbnails are stored only in this browser (IndexedDB). There’s no account. Each photo is sent to YouCam to be analysed, and nowhere else. You can export or delete everything.
Limits
Phone cameras, light and angle still add noise; split-face cancels most of it, not all. Few check-ins means wide error bars, which is why “not enough data” is a real answer. This is not medical advice. See a dermatologist for skin conditions.