Everyone has a better side. This finds yours.

It is a browser tool for finding the head angle that suits a face. One photograph can be turned in 3D, so any angle between sixty degrees left and sixty right can be sampled. Or five real photographs are captured at five fixed angles and compared side by side.

Either works. Both modes accept photographs imported from the device, and both carry a drawn demo face that runs with no camera and no files at all.

Three models run inside the browser. A face detector confirms one subject. A segmentation model cuts the person away from the background. A depth model estimates how far every pixel sits from the lens, and that depth map pushes a fine mesh forward and back. The photograph becomes a shallow relief that can be rotated.

A single photograph records only what the lens could see. Nothing exists behind the nose or beyond the cheek, so the mesh stretches the last visible pixels across surfaces that were never photographed. Up to thirty degrees the turn reads as real. At sixty it reads as an effect.

A three-quarter view, about thirty degrees off straight on, suits most faces. It gives the head visible depth, splits it into a lit side and a shadow side, and narrows the jaw slightly. Straight on reads formal. Sixty degrees approaches profile and reads dramatic.

No. Every model runs inside the browser and every photograph stays in the page. Nothing is sent anywhere, and closing the tab discards whatever was not saved to the device.

A contact sheet is one image holding every captured pose in a row, each labelled with its angle, with the starred best side outlined in amber. It downloads as a single JPEG, ready to send to a photographer or a client.

Any browser with WebGL runs it — current Chrome, Edge, Safari and Firefox all qualify. Where WebGL is unavailable the tool says so on screen instead of failing quietly, and the five-pose compare still works, because it uses no 3D at all.

The segmentation model measures what share of the picture the person occupies. Below about fifty percent there are too few pixels on the face for depth estimation to read features, and the turn goes soft. The check reports the measured figure, and the photograph can still be used anyway.

Yes. Up to five existing photographs can be imported at once. Each is checked for a single, well-framed subject and placed in the next empty angle slot. Any that fail the check can still be added deliberately.

Three angles miss the interesting middle. Thirty degrees and sixty degrees do very different things to the same face: one adds depth, the other approaches profile. Five stops make the change visible in steps, which is what lets a better side show itself.

One-shot 3D estimates angles from a single photograph, so it is fast but softens past forty degrees. Five-pose compare photographs every angle for real, so nothing is estimated. Use the first to explore and the second to decide.

The three models are fetched from a public code network the first time the studio is opened, then cached by the browser. After that the camera, the depth pass and the rendering all run locally, with no further network calls.

A few seconds on a current phone or laptop, longer on older hardware. The depth pass is given twenty-five seconds. If it has not finished by then the studio falls back to a simple estimated depth and says so in the readout, rather than leaving you waiting.

Both are mirrored, deliberately. A mirrored preview is what people are used to correcting their pose against, so the capture is mirrored the same way. The pose you see while turning your head is the pose the file holds.

The face detector counts them and stops to say so, because depth estimation on two subjects produces one confused surface. The photograph can be retaken, or used anyway if the second face is small enough not to matter.

The background is removed whenever segmentation succeeds, because a background that turns with the face smears badly. If segmentation fails the background is kept and the readout says so, which is the honest signal that edges will be rougher.

Open a captured pose and press the star. The thumbnail gains an amber ring, a best label appears over the picture, the save button then saves that pose by default, and the exported contact sheet outlines it in amber.

Yes. The studio fills one viewport and never scrolls, the picture sits at the top, and every control lives in the lower third within thumb reach. The angle dial can be swept with the same thumb that holds the phone.

Yes. A drawn demo face loads into either mode and behaves exactly like a photograph — it turns in 3D, fills all five pose slots, and exports a contact sheet. The whole tool can be learned before a single picture is taken.

A dashed line runs from your face out past the edge of the frame, ending at a pulsing dot. Look at the dot and your head arrives at the angle on its own. The dot sits outside the picture deliberately: a target drawn inside the frame can only ever turn a head about fourteen degrees, which is nowhere near enough.

The angle is measured from the face alone — where the nose sits between the eyes. Turning your shoulders with your head moves the whole picture instead of changing the geometry, so the reading stays near zero while the pose looks turned. Feet planted, shoulders square, head turns.

Yes. A face model reads your head angle several times a second. When it lands inside the target and stays there for about a second, the shutter fires on its own, then waits before it will fire again. You can always press the button instead.

Within eight degrees either side. The readout turns green when you are inside that band and the hold begins. Eight degrees is loose enough to reach without straining and tight enough that five poses still read as five clearly different angles.

Yes. A single button cycles the backdrop through black, white, cut out and off. The person is separated from the background live, in the browser, at about fourteen frames a second, and whatever you see in the preview is what the capture saves.

Cycle the backdrop to the cut-out setting and take the picture. With no backdrop behind it the file is saved as a PNG carrying real transparency instead of a JPEG, so it can be placed over any colour later without a white box around the head.

Press sweep. The five poses play in order — centre, left, right, back to centre — held long enough to read. Five still pictures side by side show you the differences; the sweep shows you the movement, and a face that suits one angle often reveals it there first.

It is a target for head and shoulders. Fill it and it turns green, then fades to a faint outline so it stops competing with the picture. If the face is too small or too large the tool says which, rather than leaving you to guess. The colour is never decoration here: it changes only when a face has actually been measured inside the shape.

Because a permission prompt arriving with no explanation deserves a no. The step before it says what the camera is for, that nothing is uploaded and that closing the tab discards everything, and it appears once. Where permission has already been given it is skipped entirely.

A switch sits in the studio bar and moves you straight across. Each mode keeps what it was holding, so poses captured in the compare are still there when you come back. Leaving by the back link returns you to the section of the article you started from.

It takes a photograph of a person, cuts their outline from the background, and fills that outline with words that are true about them. Their name sits in the middle at the largest size. What comes out is one picture, printable, made only of their own outline and their own words.

Every word traces back to a trait you picked. Each of the hundred traits carries a word set that was gathered ahead of time and stored in the page, so nothing is generated while you use it and nothing you type is sent anywhere. The machine gathered the vocabulary; you chose which of it describes this person.

Because a portrait made only of compliments is not of anyone in particular. The honest two fill a fifth of the picture at most, sit in the gaps, and are drawn in exactly the same colour as everything else. Nothing marks them out as the bad ones.

Because both halves earned it. Steady and stubborn share firm, unwavering and solid; protective and stern share the same quiet strength. A word claimed by a virtue and a fault at once accumulates size from both, which is why the middle of these portraits so often holds the truest thing about the person. Nobody wrote that in — it falls out of the words themselves.

Each trait you pick buys a fixed share of the picture, and every word inside it is sized by how strongly that trait carries it. The three you love split most of the space, the two honest ones split the rest. The name is not part of that arithmetic at all — it is set larger than whatever comes first.

No. The outline is cut in your browser, the words are placed in your browser, and the photograph never leaves the device. Closing the tab discards everything that was not downloaded.

Yes, and no occasion is the default. Around fifteen hundred occasions can be searched, or the field can be left alone. Choosing one adds the greeting to the name at the top, brings phrases people actually write on cards into the picture, and sets how much honesty belongs in it.

Yes, and the flow changes when you choose it. Occasions carrying grief are marked as ones where nothing critical belongs, and the step asking for honest traits does not appear at all — it is removed rather than shown and quietly ignored. The picture is built from the loved traits and from words of comfort.

The words are coloured from the photograph itself: the tool averages the colours inside the outline and darkens that until the type is readable, so the portrait carries the person's own light. One button flips the whole thing to dark ink on plain paper, which is what printing usually wants.

Between roughly a hundred and fifty and three hundred, depending on how much room the outline gives. The words are released over about fifteen seconds rather than appearing at once, because watching the picture assemble is most of what makes it feel like a gift instead of an export.

The tool says so and stops rather than filling the whole rectangle with words. A person standing against a busy background at a distance is the usual cause. A head-and-shoulders picture with the person clearly in front of what is behind them works nearly every time.

Yes. The camera opens inside the flow, takes a single frame with the background already removed, and hands it straight back to the portrait. It uses the same capture the compare mode uses, so the cut-out quality is identical.