Tracking a subject in a video

Tracking is how you make an effect stay on something that moves. Point at a face, a person, an object or a flat surface once, and MotherFluxer follows it frame by frame for the rest of the clip — so anything you attach to it, cut around it, or change about it travels with it.

Where tracking lives

Tracking has no menu of its own: you reach it through the thing you want to do.

• A mask's tracked items. Add a mask from the Add menu, press the target at the bottom of its card on the Mask tab, then Add tracked item and the kind you want — Track a new face…, Track a new body… and so on. It reads the clip under the playhead — or, when the mask sits over a picture you placed on the video, that picture.
• Splice, on a clip's Transform tab. Its ways to cut stand side by side in the menu's place, and Face, Body, Object and Plane each cut around what you point at — or, with Use, along something of that kind the clip already tracked, with nothing read again. Face lists a tracked person too, with why it cannot be used: a person's face carries no outline of the head to cut along, so pointing at their head reads the face on its own.
• Retouch, in the Adjust section of the selected clip's Effects strip, evens a face's skin tone: it lists the faces already tracked on the clip, and each tracked person with their Face row, each face with Use, and Add tracked item to point at a new one.
• Beauty, a section of the selected clip's Effects strip: pick a card and point at the face.
• Reframe, the reframing that keeps someone in frame, follows a subject too, and sits on the first row of the clip's Transform menu, beside its flip and Splice.

None of these is a plain add row: each has to read the footage before there is anything to set, so each has a step where you point, check and confirm. Only the things that work on that kind of subject are ever offered.

What you track is saved on the clip as a tracked object, and it stays there until you delete it. Tracking something adds no timeline row by itself; every place that uses tracking lists it the same way — a mask's tracked items, Splice, Retouch, Reframe, a mark's Follow in its Animate Recipe and an Automate lane's source — so the second thing you do to a subject never waits for a second reading. Each list shows what the item follows now with Remove, then every tracked object with how many things use it, Attach (or Use) and a trash button; pressing a row shows that object on the video. A tracked person is one row, with a Face row beneath it once their face has been read — Attach or Use on each, so a mask, a look or a reframe can take the whole person or just their face — and the trash button on the person's row alone.

Each one is named for what it is — Tracked face, Tracked body, Tracked object, Tracked plane — and the second of a kind numbers itself (Tracked face 2). The numbers are fixed when the item is made, so removing one never renames the others.

• Face — point at a face: have a mask wear a prop or face paint on it, cover it for privacy, point a callout at it, or give it a Beauty or Retouch look.
• Body — point at a person: draw their skeleton, hang a prop on a shoulder or a chest, cover them, or point a callout at them. The same run reads their face wherever it is large enough, so a person is one tracked item that also takes everything a Face does — a prop or paint on the face, a face cover, a Beauty or Retouch look — with nothing tracked twice. A person on a picture you placed is read without their face; track the face on its own there.
• Object — tap anything in the shot and follow its outline: point a callout at it, cover it for privacy, or cut it onto its own layer.
• Plane — mark the four corners of a flat surface: lay a picture onto it, or hang a prop off it, tilted to match.

Track a subject through the clip

Following something is two choices: what to follow, then what to do to it. Face, Body, Object and Plane are the four ways of saying which subject you mean.

Asking for a new track makes a new tracked object, even on a clip that already has one — so a second face, or a second object, is the same few clicks as the first. The exception is a face or a person the clip already holds: point at one and what you asked for lands on the tracked object already there, as one step, with nothing read again and no second object — and a face you point at on the head of a tracked person lands on that person's face. Tracking a person whose face the clip already holds as a face of its own takes that face in: every mask and look that followed it now follows the person's face, and the separate face goes, leaving one tracked item. An Object or a Plane is new every time. A Splice keeps what it read: the subject it cut around is saved on the clip as a tracked object, so every piece carries it and a mask on the cut-out can follow it straight away. To use something you already tracked without pointing, press Attach or Use on its row in any tracked items list. A tracked object goes only when you say so: its row says how many things use it — masks and looks alike, used by 1 item — and its trash button deletes it and lets every mask following it go. A Beauty look that followed a deleted face goes with it, since it would draw nothing; any other look stays, and a Retouch then evens the skin tone across the whole clip. It is one step, which Undo brings back. Releasing every mask that followed it leaves it saved.

From Retouch, the menu shows one line telling you what to do; from Splice, that line shows on the clip itself instead of in the menu, and for an Object it goes once you have tapped, so nothing covers the outline while you check it. Either way the buttons stand in a small container just under the video (or above it, when the menu along the bottom would cover it). They are icons, with the one that finishes highlighted. For an Object, Add (+) and Subtract (−) come first, with a hint saying what your next tap does — Tap to add or Tap to remove — then Undo, which steps your selection back one tap at a time, Cancel, and the highlighted Done: Track when you are tracking, Splice when you are cutting. A face or a body takes a single tap, so it has just Cancel and Done. What was found is drawn on the video so you can see it before anything else happens. Done stays dim until something is found, and pressing it is what starts the run; Cancel closes the flow with nothing created. While the tracking runs, Tracking… and how far along it is show on the clip, and Cancel in the menu stops it. From a mask, the same container appears while you point, and a note at the mask reads Tap the thing to track.; Reading the picture… shows on the mask while it runs, and the ✕ at the top of the tracked items list stops it.

• Face and Body look for your subject before asking you anything. Open one and it reads Finding subjects… while it searches — the frame you are on first, then a spread of frames across the clip — and draws what it found on the video, each inside a light box you can press to pick it. Find exactly one and it is chosen for you, so Track is ready to press immediately. Find several and the menu says how many and offers them as chips to pick from. Faces are found at any distance, so a face far back in a wide shot counts as much as one filling the frame. A person's outline takes in the whole person, head included, wherever you tap them, and a face's takes in the whole head with its hair, not only the skin.
• Tapping still works, and it always wins. Point anywhere on the video to track someone that was not offered, or to change your mind after a pick. What you tap is what gets followed.
• Object is a tap from the start. Tap the thing you mean, anywhere in the shot, or drag a box around it, and its outline appears; refine it with Add and Subtract if it caught too little or too much. It is not limited to a list of things a computer was taught to recognize.
• Plane is the one that takes a rectangle instead of a point. Click once to drop the first corner, and the rectangle follows your mouse; click again to finish it, then drag any corner so it sits on the surface. Those corners are what the effect follows — and they carry straight through to laying an image or a prop on that surface, so you never place them twice.

The video stays exactly where it is for the whole of this — the frame you were looking at when you started is the frame you are pointing at, and nothing starts playing while you choose. You point at the clip where the display draws it, moved, resized or tilted as it is, and your point lands on the right part of the footage. What gets read is the footage itself, without the looks on the clip, so an effect never changes what the tracker sees. For a step you start from the clip, Expand it first from the ⋯ More menu to point at it alone, at its own shape, without its placement or its group's movement. Pausing in the middle is fine and often better — tracking runs outward in both directions from the frame you chose, so the frames before it are followed just as well as the ones after.

If you press Undo while a track is being worked out, it waits for the track to finish, then takes it back.

A subject that passes behind something and comes back out is picked up again as the same subject rather than becoming a second one — it is matched on where it went out of sight, how fast it was moving, and what it looked like, so a second person standing beside the thing it walked behind doesn't inherit its effects.

Tracking keeps to the shot. An uploaded video can hold several shots joined by hard cuts, so before following anything MotherFluxer finds where the picture cuts. An object — and the outline of any face or person you track — is followed only through the shot you pointed at it in, up to the cut and no further: across a cut it stops rather than latching onto whatever in the next shot happens to look like it, and a face's or a body's mark jumps at the cut instead of sweeping from one person to another. A quick flash or a single odd frame is not taken for a cut. To follow the same subject in another shot, scrub into that shot and correct the track there with Fix the tracking here. Each shot's outline is worked out only from what you pointed at in that shot, so a correction in one shot never leaks into another.

When it can't find your subject

One unlucky frame is not the answer to "is my subject in this clip", so pointing at a subject searches wider than the frame in front of you: the frame you are on, then a closer look right where you pointed, then the frames around it, then a spread across the whole clip. Only when all of those come back empty are you told nothing was found — so a "nothing found" message is a real answer, and which message you get tells you what to do next:

• Nothing found there — select somewhere else. Frames were read and held nothing of that kind where you pointed. Point somewhere else; the clip itself is fine.
• Nothing found on this frame — press Track anyway and the whole clip gets searched. Only the frame on screen came back empty. Pressing Track is often the right answer.
• Couldn't read the video at this moment. The frame itself could not be read, so nothing was ever examined. Move to another part of the clip — re-pointing on this one won't help.
• This clip is still being optimized. An uploaded clip is still being prepared. Wait a moment and try again; this one clears itself.
• This face is too small or turned away for this effect. The face was found and can be followed, but its features can't be read. A mask that sits on those features and the Beauty looks can't draw on it; a cover or a blur still can. Choose a larger, forward-facing face.
• Tracking couldn't start on this device. Nothing started up, so no frame was looked at. Reload the page and try again.
• Tracking stopped partway — this device's graphics reset. The tracker had started working through the clip when the device's graphics restarted, so that run stopped. Reload the page and track again.
• This clip couldn't be prepared. Press Try again on the clip, then try again here. The clip's own working copy failed to be made, so no frame could be examined, and waiting will not fix it. The clip shows the failure with its own Try again; press that first, and point again once the clip is ready.

There is no separate tracking tool to open — you always start from the thing you want to do, and it already knows which kind of subject it needs. The depth effects have no subject step, because they read how far away everything is rather than following one thing; captions have none either, and they are made on the Add tab rather than here.

The two reframing effects are the one place a subject can be either kind — they offer a switch between following a face and following a person.

Tracking an object says this device can't run it

Effects that follow a specific thing you tap need a recent browser or device to work out its outline frame by frame. When yours can't, MotherFluxer says so the moment you open the effect, instead of making you wait for an analysis that was never going to finish.

• Update your browser to its current version and try again — this is the usual fix.
• Try the same Video on a newer device if updating doesn't clear it.
• Effects you already tracked are unaffected. They keep playing, rendering, and exporting on any device; only creating a new one is blocked.
• Effects that find their own subject still work — anything based on faces or scene depth doesn't ask you to tap.

Correct a track that went wrong

Tracking runs ONCE per subject, and it reads everything about that subject in that one pass — where it is frame by frame, and its exact outline. So the first use and the tenth cost the same: point a second mask at a face, or give it a Beauty look, long after you tracked it and it appears straight away, with no second wait. A Beauty or Retouch look applies the moment you pick it, and its Recipe changes it live. The one confirmation in the whole flow is the one that starts the tracking itself, and it is there so you can see what was found before you wait for it.

If a track goes wrong partway through — it latches onto the wrong person as they cross, or the outline slides off as the subject turns — you don't have to throw it away and start over. In the tracked items list of a mask following it, press the object's row; under it, scrub to a frame that looks wrong, press Fix the tracking here, and point at the subject again on that frame. Where the tracker was unsure, the same place also offers Fix the weak tracking at a moment: press it and the video jumps to the weakest frame of that stretch with the correction already waiting for you to point. The Recipe of a look that follows a face offers the same two in place of its Region section, and the tracking panel offers them when it opens on a subject you already tracked. A mask on a picture has neither, because a picture is read once.

A correction is pointed exactly the way the first selection was, on the frame you are on — the video stays on that frame, and nothing is searched for first. A note at the clip, or at the mask, reads Point at the subject on this frame, or Correct the outline on this frame for an object or a plane, and the same buttons stand in the small container under the video. For an object that is Add (+) and Subtract (−), with the hint saying what your next tap does — Tap to add or Tap to remove — so you can tap, or drag a box around the subject, and keep refining the outline; Undo steps it back one tap at a time; then Cancel and the highlighted Track. A face or a body takes a single tap, so it has just Cancel and Track. Cancel leaves the track as it was, Escape ends the correction before it closes anything else, and closing the menu does not stop a correction that is already running.

What you just said joins everything you have already told the tracker, and the whole clip is worked out again from all of it together. So a correction improves the frames before it as well as the ones after, and an earlier correction is never undone by a later one. The trade is time: a correction re-solves the clip, so it costs about the same wait as the original track rather than a fraction of it. Each correction is remembered as part of the track, so anything that re-reads the subject later never quietly goes back to the version you rejected.

See what you tracked

Every object you have tracked on a clip is drawn on the video while that clip is selected, following it frame by frame — so you can see exactly what the tracker is holding on to, and watch whether it holds all the way through the shot. The one the selected mask follows is drawn brightest and the rest stay faint, so the picture always tells you which one it is — and pressing a row in a tracked items list draws that one brightest instead. A mask that follows something has no selection box of its own, so this mark is how it shows you where it is. Only the mark is drawn — the names live in the menus, where reading them costs you none of the picture.

You see what was actually captured, as one clean figure per tracked object — never a cloud of points, and never a rectangle around it:

• A face shows a head in 3D, softly lit and set exactly on the points the tracker found, so it turns, tilts and moves with the real head.
• A body shows a whole figure standing on the joints the tracker found — one skin from head to feet that bends at every joint as the person moves, with their face in its head when the tracker read one. A joint the tracker could not see leaves its part of the figure faint.
• A surface shows the plane it found: its four edges, a grid inside them drawn in the surface's own perspective, and a bracket at each corner — so you can see the shape and the angle it will map onto rather than the box it sits in.
• An object shows its traced silhouette, lightly filled — the exact edge a cut-out would follow.

Some of this is read from files made when you tracked, so it fills in a moment later: a face shows its box until its head lands, a person's figure gains their face in its head the same way, and an object shows the tracker's box until its silhouette arrives, each upgrading in place with nothing to wait for.

Marks are drawn only while that clip, or a mask following one of its objects, is selected, and they are never part of your exported video.

Correcting the tracking from the mark. Open a tracked face or body and a few small rings mark its key points — a face's two eyes, nose tip and chin, a body's twelve joints. Drag a ring to where that point really is on this frame and the tracker re-solves the whole clip from your correction; with the keyboard, focus the mark, nudge the ring with the arrow keys (Space moves to the next point), and press Enter. Escape puts a nudged ring back. If nothing is found where you dropped it, the note at the mark says so — tap the subject on the frame instead and press Track, exactly as a correction from the menu goes; while a correction is still running, the rings wait.

Pick what to do to your subject

What you can do to a subject depends on what it is, and only what the tracker can actually deliver for it is offered. Tapping a chair never offers you lipstick: a mask card the subject cannot carry is shown dimmed and says what it needs, and a Beauty card always asks for a face.

Masks — put something on it, and keep it there as it moves. Each of these is a style a mask draws, picked from the strip of cards on the mask's Mask tab while it follows the subject:

A mask's settings live on the mask's own Mask tab: the container at its right shows the mask, and the settings button in the container's bottom row, beside the tracking target, opens the Recipe of the style it wears.

• Wear — a ready-made lens from the library — a prop, face paint, or both — worn on the kind of subject it was made for: a face, a body, or a marked surface. Each card shows its prop straight on, as it looks when the subject faces the camera, so you can tell them apart before you pick one.
• Privacy — Blur, Pixelate, Bar or Eye bar, over anything with an outline.
• Callouts — Circle hugs the subject and resizes to it every frame, Underline stays under it, Pointer re-aims at it from a fixed point, Label ends its leader line on it, and Trail draws a tapered trail of its recent path, optionally coloured by speed.
• Skeleton — Skeleton, Pose trail, Form and Ghost, drawn from a person's joints. People only.
• Surface — Replace and Decal, a picture mapped onto a marked surface. Marked surfaces only.

Splice — separate it into its own layers

Splice does not treat a subject, it CUTS one out: it lifts what you pointed at onto its own layer above the original, either keeping it in the original as well or leaving a see-through hole exactly where it was. Because each piece is its own clip, you can move, resize, retime or add any effect to it on its own — so "blur the background" is a splice plus an ordinary blur on the background piece, and every effect the editor has is available to each piece rather than a short list of built-in ways to treat one side of a silhouette.

Splice can also cut by DISTANCE rather than by subject: it reads how far away each part of the shot is and separates the picture into layers, near to far. Those layers arrive as a group, and a Depth look on the group then lights them, fogs them, grades them and defocuses them by how far away each one is.

Privacy covers are masks. Add a mask, pick a card from its Privacy section, and have it follow a face or an object; it covers what it follows. Unlike a blur chosen for looks, a cover never uncovers on a frame the tracking is unsure about: it grows to cover more, and if it loses the subject altogether it covers the whole picture rather than nothing. It takes no effects, border or fade, because a cover at half strength would show what it hides.

Make a mask pulse with the music. While a mask follows a subject in a video with sound, its Tracking list also offers Pulse size, Throb opacity and a glow. Switch one on and the mask swells, fades or glows on the beat while it stays on its subject; React to picks the part of the sound it listens to, and turning every one off stops the pulse. The swell and the fade are the mask's Beat movement, so they also show in its Animate Recipe, where they are tuned or removed like any movement. A privacy cover has none of these, because a cover that faded on the off-beat would uncover what it hides.

Expand a mask to look at it alone. Expand, in the ⋯ More menu of the row above a selected mask, fills the picture area with the mask on its own. A mask that follows something stands exactly as it sat at that moment, held still, even on a clip that moves: its group's movement is left out. Back to the whole video returns, with the mask and what it follows unchanged.

Beauty — change the face's own appearance. The Beauty looks are a face's retouches. The Beauty section of the Effects strip holds Smooth, Brighten, Bigger Eyes, Slim Face, Lips, Eyes, Brows, Blush and Glam. Pick a card and nothing changes until you point at the face; then the look lands on that face and follows it, all as one step Undo takes back, and Cancel leaves nothing behind. Pointing at a face the clip already tracks — a tracked person's face included — uses it again with no second wait, and picking another card on a layer that follows a face switches the look in place, with no pointing at all. Each look is an ordinary effect layer on the face it follows, with its colour and amount in its Recipe, and it draws live, moving with the face continuously as the clip plays. Retouch, in the Adjust section, evens the skin tone.

Mark a surface and lay something on it

The Plane starting point is for flat things in the shot — a wall, a screen, a laptop lid, a table top, a sign. Instead of tapping, you place four corners onto the surface on the video, and those corners are the whole input: from them MotherFluxer works out both where the surface is and which way it faces, frame by frame.

That gives you two things a plain outline cannot:

• A mask's Surface styles: Replace swaps it for an image you upload, and Decal lays a see-through graphic over it so the real surface still shows through. Either one tilts and turns with the surface as the shot moves.
• A lens made for surfaces, from a mask's Wear section, hangs a ready-made prop off the surface — a sign on a wall, a decal on a car door — angled to match the way the surface faces rather than pasted on flat. It can be worn only while its mask follows a marked surface.

The surface is tracked through the shot, so it holds when the camera or the surface moves. If the surface is nearly square-on to the camera, or the shot is too shaky to be sure which way it faces, the effect keeps the plain straight-on view rather than guessing at an angle.

Every frame is measured against the frame where you placed the corners, not against the frame before it, so the surface doesn't creep off the wall over a long shot the way a running estimate would. When something passes in front of it, the lock returns as soon as the surface is visible again.

Props take their angle from what they hang on

A prop is not pasted onto the frame — it takes both its position and its angle from the thing it hangs off, so it turns as that thing turns.

• On a face, a prop follows the head: glasses stay on the nose through a head turn, a crown stays level with the brow, and a prop keeps its size as the head turns. A crown, goggles or a mask that sits over the face is drawn in front of it, and one worn behind the person, such as wings, stays behind them, cut by the whole head and body. Behind subject, in the lens's settings, changes which it does. Sway, in the same settings, lets a prop trail its subject a little, from welded on at 0 to an exaggerated swing, and Lag beneath it — its Period and Bounce — sets how that swing moves; it belongs to the subject, so every prop on it swings alike.
• On a body, a prop follows the part it is pinned to. A chest emblem turns with the chest, a shoulder prop with the shoulder — the body's parts each carry their own facing, so a cape does not stay flat while its wearer turns away.
• On a marked surface, a prop follows the surface's own facing, so a sign on a wall angles with the wall.

The pose behind this is a good read on the trunk of the body — chest, hips and head. Hands, forearms and legs are not reliable enough to hang a prop on, so nothing is offered there.

Fine-tune what an outline covers

Following an Object — to cut it out with Splice, or to hang a callout or a privacy cover on it — traces an outline around the thing you tapped. Your first tap is a starting point, not the final answer, so you can correct it before anything is analyzed.

• Tap the thing you want, or drag a box around it. The outline appears on the frame within a second or so.
• Refine it with Add and Subtract. With Add (+) on, the hint reads Tap to add and your next tap brings more into the outline; with Subtract (−) it reads Tap to remove and your next tap takes something out. Each tap re-reads the whole selection, so it keeps getting closer, and Undo steps it back one tap at a time.
• Then track it. The outline you approved is exactly what gets followed through the clip.

The outline follows the subject's real shape, not just its position: as it turns, moves closer, or changes shape, the cut re-traces it frame by frame instead of sliding one fixed silhouette around. That is what keeps a cut-out tight against a person who turns around, or a cover on an object that rotates.

Reframe the shot around your subject

Reframe sits on the first row of the clip's Transform menu, beside its flip and Splice, and reads what the clip carries — the effect's name, or Off. Press it and its two effects — the two that move the camera rather than marking the subject — stand across the top of the menu, the one the clip carries already lit and open; on a clip with no reframe yet, pressing Reframe starts no work until you pick one. Both follow a subject, and a clip you have already tracked has one: once you pick, a row under them offers each of your tracked items by name, or a fresh Track object:

• Reframe — a virtual camera that keeps your subject in frame, great for turning a wide shot into a vertical one.
• Punch-in — a zoom that pushes in on and follows your subject.

Both follow a person or a face. They start on whichever suits the effect, and offer Or track a face / or track a person underneath if your shot suits the other one better. A person tracked with their face is both at once: press Attach on the person's row to frame the whole person, or on their Face row to frame just the face, and switch between the two with nothing read again.

Depth

Depth effects read how far away each part of the shot is instead of following one subject, so they have no step where you point. There are two ways to use them.

A Depth look on the clip. Pick a card from the Depth section of the clip's Effects strip. The clip's distances are read for you, with N% complete on the clip while they are, and the look treats the picture by how far away each part is: fog that thickens with distance, a lamp lit against the scene's own shape, a stage light, a tilt-shift defocus, god rays that break around anything nearer, red/cyan 3D, and a warm golden-hour haze that sinks into the far field.

A splice into layers. Under Splice on a clip's Transform tab, choosing Depth separates the picture into layers, near to far, as a group of clips. What you then do to those layers is ordinary editing: each one moves, resizes and retimes on its own and takes any effect the editor has. A Depth look on the group treats them all together, and the first thing in its Recipe is the layers on one Near-to-Far line, each an exact distance you can drag.