Pins and solving

Adding photos

Bring photos in from your project's assets, or import them straight from disk, in the left panel. Each one runs through automatic face detection on add, which places a starter set of pins and gets an initial pose. If detection can't find a face, the photo is still added; you pin it by hand from there.

Each photo's row shows its solve status: a reprojection error in pixels once it's solved (green under 8px, amber above), or unsolved if it hasn't reached the 6-pin minimum yet.

Two re-detect buttons sit on each row:

Re-detect landmarks keeps any pin you've dragged, and only refreshes the ones you haven't touched.
Full re-detect rebuilds every automatic pin from scratch, including ones you've moved, and re-projects the anchor guides fresh. Pins you placed by hand are the one thing it always keeps, since those aren't something detection can regenerate.

The four kinds of pin

Pins are color-coded by what part of the head they belong to, and split into four kinds by where they came from:

Automatic pins come from face detection, on the roughly 44 landmarks it tracks across the face oval, eyes, brows, nose, and mouth. These are what re-detect manages.
Anchors cover the parts of the head detection can't see at all: ears, skull, neck, shoulders, chest. Kitly knows where each of these sits on the underlying head shape; you place where it falls in each photo by hand. These matter more than they look like they should: the head shape moves far more at the neck and skull than it does on the face itself, so a couple of well-placed anchor pins do a lot of work.
Named points are a manual pin you've explicitly named, so the same physical spot (a mole, a scar) gets recognized as the same point across every photo instead of being treated as an unrelated pin on each one.
Manual pins are a plain click-to-place pin, with no guarantee it's the same spot from photo to photo.

Guides versus confirmed pins

A hollow ring is a guide: Kitly's best guess at where an anchor or automatic pin should sit, based on the current fit. It's not yet feeding the solve. Drag it onto the real feature and it fills in solid, confirming it. A pin you've deliberately turned off shows as a hollow grey ring instead, so a guide and a disabled pin never look the same.

Placing an anchor doesn't need the photo's wireframe to actually cover that part of the head. Click where the anchor point is and Kitly places it there directly, which is what makes it possible to pin the back of the head from a three-quarter shot.

On-canvas controls

Wireframe overlay opacity: how visible the projected head mesh is over the photo. Display only.
Residual-error threshold: rings any pin whose own reprojection error is above the slider, so you can spot a bad pin rather than reading the sequence's overall error number. Also display only; it never changes the solve.

Solving

Once your pins are placed, the Solve section in the inspector runs the fit:

Regularization: how strongly the head is pulled toward a neutral shape. Higher keeps it closer to generic; lower lets it move further to match your pins.
Iterations: how many rounds of pose-and-shape refinement to run.
Re-sweep focal each round: on by default. Without it, a bad focal-length guess from early in the solve can stick around uncorrected.
Fit per-photo expression: off by default. Useful for a candid photo set where the expression drifts a little between shots; wasted effort on a controlled, neutral shoot.
Symmetry: how strongly the two sides of the head are pulled toward matching each other. Turn this down for a genuinely asymmetric face.

Cleaning up back-facing pins

Face detection reports all its landmarks even on a strong profile shot, so roughly half of them are guessing at a side of the face the camera never actually saw. Disable back-facing pins, in the selected photo's section of the inspector, turns those off for you. It only touches automatic pins you haven't dragged yourself, nothing is deleted, and Ctrl+Z undoes it.