Why photos from your gallery don’t work
Skin analysis reads small differences in color, shadow and surface detail. A selfie by a window in the morning and one under a bathroom lamp at night differ in exactly those things, so their scores differ even when your skin has not changed. Distance changes how many pixels each pore gets. A head turned slightly to one side puts one cheek in shadow and flattens the other. Two photos from your gallery mostly compare two moments of light.
So Kept takes every photo itself and checks the same things each time. The camera measures your face in the live video and waits until all of these pass. When they stay green for a moment, it takes the photo on its own.
| Check | Passes when | Otherwise the camera says |
|---|---|---|
| Face size | Your face spans 55% to 80% of the guide’s width | Move closerMove back |
| Angle | Head turned or tipped up and down by 10° at most, tilted to the side by 8° at most | Look straight at the cameraTilt your chin down a little |
| Light | Forehead and cheeks average 90 to 200 on a 0 to 255 brightness scale, and the two cheeks differ by 35 or less | Find more lightToo bright, step away from the windowLight is coming from one side |
| Line up, check-ins only | Twelve points on your face (eye corners, nose, mouth corners, chin, forehead, cheek edges) sit on average within 10% of the distance between your outer eye corners from where they were on day 0. That is about 1 cm at arm’s length | Match the outline from day 0 |
| Sharpness, uploads only | At least 40 on Kept’s sharpness measure. A sharp phone photo scores about 160 to 185, the same photo blurred by 1 px about 60 | Hold still or use a sharper photo |
At check-ins the oval on screen is replaced by a faint outline of your day 0 face, so you stand where you stood the first time. Uploaded photos go through the same checks plus sharpness, because a blurred photo hides pores and texture.
Some differences still get through. If a check-in photo is 10% or more lighter or darker overall than day 0, the verdict says so and its confidence drops to medium. Past 18% it drops to low. A check-in that sits further from the day 0 outline than the camera allows also lowers confidence to medium.
What YouCam measures
Kept uploads each photo through the YouCam File API and sends it to AI Skin Analysis, the SD model, asking for up to 8 skin areas: the ones the product targets and the ones worth watching for irritation. For every area YouCam returns a score from 1 to 100, where higher always means healthier looking skin, and a map of where on the face it found the problem.
Kept keeps both. On the acne map it counts separate spots, which is the blemish count you see in verdicts. On the redness map it measures the share of the face that is red. These are the areas Kept works with:
| Area | Ingredients that target it | In the target photo |
|---|---|---|
| Acne | Salicylic acid, Benzoyl peroxide, Adapalene, Retinol, Retinal, Azelaic acid | Yes |
| Redness | Azelaic acid, Niacinamide, Ceramides and hydrators, Soothing extracts | Yes |
| Pores | Salicylic acid, Adapalene, Niacinamide | Yes |
| Texture | Adapalene, Retinol, Retinal, Gentle retinoids, Glycolic acid, Lactic and mandelic acids, Ceramides and hydrators | Yes |
| Fine lines | Retinol, Retinal, Gentle retinoids, Peptides | Yes |
| Hydration | Ceramides and hydrators | No, YouCam can’t simulate it |
| Oiliness | Salicylic acid, Niacinamide | Yes |
| Radiance | Glycolic acid, Lactic and mandelic acids, Vitamin C, Tranexamic acid and arbutin | Yes |
| Dark spots | Retinol, Retinal, Glycolic acid, Azelaic acid, Niacinamide, Vitamin C, Tranexamic acid and arbutin | Yes |
| Dark circles | Caffeine | Yes |
| Eye bags | Caffeine | Yes |
| Firmness | Peptides | No, YouCam can’t simulate it |
The target photo
When a trial starts, Kept sends your day 0 photo to AI Skin Simulation with the areas the product targets and a strength for each, set per ingredient. Salicylic acid, for example, asks for acne at 1, pores at 0.7 and oiliness at 0.7. YouCam returns your face with those areas improved. Kept then analyzes the simulated photo like any other, and its scores become the end point of the expected curve.
The curve stays flat until half of the ingredient’s first change day, rises along a smooth S shape and reaches the target scores on the full effect day. The target photo is a reasonable picture of success for this product on your face. It is not a promise.
Sunscreen
Sunscreen prevents damage that builds up over years, so a trial of a few weeks can’t show it working. For a sunscreen Kept skips the trial and uses YouCam AI Aging Simulation to show how your face may age instead.
How much change is real
Analyzing the same photo file twice gives identical scores, so YouCam adds no randomness of its own. The spread comes from the photo. Two photos taken a minute apart are never quite the same: you stand a little closer, the light shifts, your head tilts a degree.
To measure how much that moves the scores, we took one face and made 4 variants of it, each a change small enough to pass the camera checks. Then all 5 were analyzed with the SD model.
| Photo | Acne | Redness | Pores | Texture | Oiliness | Hydration |
|---|---|---|---|---|---|---|
| Original photo | 66 | 64 | 69 | 74 | 76 | 67 |
| Moved 3% across and 1% down, scaled to 97% | 66 | 63 | 66 | 75 | 77 | 67 |
| 6% brighter | 66 | 64 | 68 | 74 | 76 | 67 |
| 6% darker, rotated 1.5° | 66 | 64 | 70 | 73 | 79 | 68 |
| Blurred by 0.6 px | 66 | 64 | 69 | 77 | 77 | 68 |
| Highest minus lowest | 0 | 1 | 4 | 4 | 3 | 1 |
| Twice the standard deviation | 0.00 | 0.89 | 3.03 | 3.03 | 2.45 | 1.10 |
Kept takes twice the standard deviation of each area, picks the largest among acne, redness, pores and texture and rounds it up: 3.03 becomes 4. Oiliness and hydration, measured on the same photos, would not change that number: their largest is 2.45.
A change of 4 points or less from day 0 counts as no change. More than 4 counts as a real change. More than 8 counts as a strong one, which is what Kept looks for when deciding whether your skin is reacting.
Every verdict compares your latest check-in with day 0, and the same number appears on the photo, the chart, the verdict and the report. One unlucky photo is caught by the camera checks before it is taken, and by the confidence note when the light differs from day 0.
This was measured on variants of one photo, not on photos of a person taken on different days. Real repeat photos can differ more, for example after a short night. That is why the camera checks matter, and why the verdict lowers its confidence when light or framing drifted.
Timelines by ingredient
Every ingredient in Kept’s catalog has its own timeline: when the first changes usually appear, when the full effect is reached and, for some, a window when a short breakout is common because skin renews faster. Before the first change day Kept says it is too early to tell, unless your skin is reacting. Inside the breakout window a rise in acne is read as likely purging rather than failure.
| Ingredient | What Kept tracks | First changes | Full effect | Breakout window |
|---|---|---|---|---|
| Salicylic acid | Acne, pores and oilinessWatches redness | 2 weeks | 8 weeks | Days 7 to 42 |
| Benzoyl peroxide | AcneWatches redness and hydration | 2 weeks | 8 weeks | Not expected |
| Adapalene | Acne, texture and poresWatches redness and hydration | 3 weeks | 12 weeks | Days 14 to 49 |
| Retinol | Texture, fine lines, dark spots and acneWatches redness and hydration | 4 weeks | 12 weeks | Days 14 to 42 |
| Retinal | Texture, fine lines, dark spots and acneWatches redness and hydration | 3 weeks | 12 weeks | Days 14 to 42 |
| Gentle retinoids | Texture and fine linesWatches redness | 4 weeks | 12 weeks | Not expected |
| Glycolic acid | Texture, radiance and dark spotsWatches redness | 2 weeks | 6 weeks | Days 7 to 28 |
| Lactic and mandelic acids | Texture and radianceWatches redness and hydration | 2 weeks | 6 weeks | Not expected |
| Azelaic acid | Redness, acne and dark spotsWatches hydration | 4 weeks | 12 weeks | Not expected |
| Niacinamide | Pores, oiliness, redness and dark spotsWatches hydration | 4 weeks | 8 weeks | Not expected |
| Vitamin C | Radiance and dark spotsWatches redness | 4 weeks | 12 weeks | Not expected |
| Tranexamic acid and arbutin | Dark spots and radiance | 6 weeks | 12 weeks | Not expected |
| Ceramides and hydrators | Hydration, texture and redness | 1 week | 4 weeks | Not expected |
| Soothing extracts | Redness | 1 week | 4 weeks | Not expected |
| Peptides | Fine lines and firmness | 8 weeks | 12 weeks | Not expected |
| Caffeine | Eye bags and dark circles | 2 weeks | 8 weeks | Not expected |
| Sunscreen filters | No trial. Its effect shows over years, so Kept shows an aging simulation instead | |||
When a product contains several of these, the trial uses the earliest first change day, the latest full effect day and one breakout window that covers all of them. Check-ins fall on days 7, 14, 21, 28, 42 and 56 up to the full effect day, with the last one on the full effect day itself. These are typical ranges for each ingredient, not a prediction for your skin.
How the verdict is decided
The verdict is computed in your browser by fixed rules from the scores, the day of the trial and the ingredient’s timeline. Kept checks the rules in this order and stops at the first one that applies.
Your skin is reacting
Redness or hydration, when Kept watches them, is more than 4 points worse than day 0 outside a breakout window, or more than 8 points worse in the latest photo at any time. If it was already worse at the check-in before, the next step is to stop. Otherwise Kept suggests using the product every other night for a week.
Likely purging
The latest check-in falls inside the breakout window, acne is more than 4 points worse than day 0, and redness has not dropped by more than 8. The next step is to keep going on schedule.
Too early to tell
There is no check-in yet, or the latest one comes before the day the ingredient usually shows first changes.
Working
The target areas improved by 4 points or more on average, and at least one of them by more than 4.
Not doing enough
The full effect day has passed and the target areas did not improve that much.
On track
Anything else. Kept compares the latest scores with the expected curve and says whether you are ahead of it or behind it.
Each verdict lists up to 4 reasons with the numbers behind them, such as the score on day 0 and today. Confidence drops when light or framing drifted, and drops to low when you started another product during the trial, because changes after that may come from it.
An optional assistant, built on Gemini through OpenRouter, answers questions about a trial. It reads the trial’s numbers and explains the verdict. It never decides it.
What stays on your device
There is no account. Your trials, photos, skin maps, scores and diary notes are stored in this browser’s IndexedDB, on this device.
To analyze a photo, Kept’s server forwards it to the YouCam API and passes the result back. The server does not save photos or results. YouCam’s links to the maps expire after two hours, so the browser downloads them as soon as an analysis finishes.
The assistant receives the trial’s numbers and diary notes, without photos. Photos go to OpenRouter only if you switch on “Let the assistant look at my photos”, and then only two of them, reduced to 512 px.
Because everything lives in the browser, clearing this site’s data deletes your trials, and they do not follow you to another device. The report page of each trial can be printed or saved as a PDF.
Limits
Kept compares photos of your skin. It is not a medical diagnosis, and a score from 1 to 100 says nothing about what causes the thing it measures.
The ±4 point range was measured on variants of a single photo. Sleep, your period, sun or a new product can move scores further from one week to the next. The check-in diary lets you note these, and a new product lowers the verdict’s confidence.
The timelines are typical for each ingredient. Your skin can respond faster or slower.
See a doctor rather than waiting for the next check-in if you have pain or swelling, burning that does not settle after you stop the product, a rash that spreads, or a spot or mole that changes in size, shape or color, bleeds or does not heal.