← PicsartCONTENT HISTORY

Update to Picsart

Snapshot Nov 7, 2026 · 00:09 UTC · version 2.4.0

Collection source: downloaded plugin package.

WHAT CHANGED

Instructions updated for motion-tracking-from-video

Skill instructions — Now says: metadata: ported-from: motion-extraction unversioned at 55.75.0 - **A track that is covered, then reacquired.** After the confidence collapses, the later...

Observed in instructions or declared skills. Runtime behavior has not been tested.

Save Picsart and check future changes in Following.

Up to 3 plugins free · no card required.

All fields and original values.

Skill instructions

Before

Nov 3, 2026 · 18:14 UTC

No removed lines (ordering may have changed)

After

Nov 7, 2026 · 00:09 UTC

metadata: ported-from: motion-extraction unversioned at 55.75.0 - **A track that is covered, then reacquired.** After the confidence collapses, the later samples can be a valid track of the object that replaced it (a card becoming a ...

Compare saved observations

Download comparison JSON

SKILL.md text changes

− removed · + added

--- before
+++ after
@@ -1,6 +1,8 @@
 ---
 name: motion-tracking-from-video
 description: Measure what objects visibly do in a video - where each one moves, turns and grows, frame by frame - with picsart_media_track_motion, and turn the measurement into scene keyframes; then measure your own render the same way and compare the two with picsart_media_track_motion_diff. Covers choosing targets and estimators, tolerances, reading the result honestly, and placing tracked motion on a layer. Use when a scene must reproduce motion seen in a reference video, or when a render's motion must be checked against one.
+metadata:
+  ported-from: motion-extraction unversioned at 55.75.0
 ---
 
 # Motion extraction
@@ -90,6 +92,13 @@
   are biased. Start once it is fully on screen.
 - **Clip the window, not the keyframes.** A fit over frozen samples emits false keyframes at the
   tail. Track again with an earlier `t1`; never prune the keyframes by hand.
+- **A track that is covered, then reacquired.** After the confidence collapses, the later samples can
+  be a valid track of the object that replaced it (a card becoming a chip). Check them against a
+  still before discarding them.
+- **One object, two stretches that need different estimators** (a video inside a card, then siblings
+  touching it): track each stretch on its own, with one overlapping frame. The two estimators'
+  reference points differ by about 2 px: measure that offset on the overlap frame and subtract it
+  when you join them, and multiply the second stretch's scale by the overlap frame's scale.
 
 | Content | `mode` | `estimator` |
 |---|---|---|
@@ -153,3 +162,10 @@
   a mismatch.
 - **What the deviation says.** A constant offset is geometry: a position or a size. A deviation
   that grows and shrinks along a move is easing. One that peaks at a beat's start or end is timing.
+
+## Measuring for a seam
+
+When a seam decision depends on a reference's velocity, path or overlap, the measurement feeds
+`motion-scene-continuity` (fetch with `picsart_scene_recipe`). Tracking answers what moved, where
+and when; continuity judges what that movement communicates and whether the scene should preserve,
+replace, interrupt or reset it. A close numeric fit does not by itself prove a readable handoff.
Full technical diff · 1 changed fields

changed /skill_md_contents

BEFORE
"---\nname: motion-tracking-from-video\ndescription: Measure what objects visibly do in a video - where each one moves, turns and grows, frame by frame - with picsart_media_track_motion, and turn the measurement into scene keyframes; then measure your own render the same way and compare the two with picsart_media_track_motion_diff. Covers choosing targets and estimators, tolerances, reading the result honestly, and placing tracked motion on a layer. Use when a scene must reproduce motion seen in a reference video, or when a render's motion must be checked against one.\n---\n\n# Motion extraction\n\n`picsart_media_track_motion` recovers what objects **visibly do** in a flattened video: their\nposition, rotation and scale on every frame, fitted into scene keyframes. It does not recover the\nkeyframes someone authored. It produces new motion data that reproduces what is on screen.\n\nLooking decides **what** to track. The tracker measures **where**, every frame, to a fraction of a\npixel. Never write a motion's numbers from a still or from memory: every timing and distance comes\nfrom a tracking result.\n\n## Cost and receipts\n\nNeither `picsart_media_track_motion` nor `picsart_media_track_motion_diff` is a SPEND unless its\ndescription says it charges (picsart-video-rules), so track as often as a beat needs. Each replies\nwith a job receipt by default: follow it as picsart-video-rules says. Targets that share a tracking window go in one call, as many as\nits `targets` schema allows: one `t0`/`t1` holds for every target, and each must be fully visible at\n`t0`. A target that enters later, or is covered before `t1`, gets a call with its own window.\n\n## The method\n\n1. **Look at the first frame of the window** with `picsart_media_video_stills`, and choose what to\n   track. On a flat background, `{ \"auto\": true, \"max\": 9 }` finds the distinct objects at `t0` and\n   tracks each, ranked by how much they move. Objects that overlap merge into one, so give those a\n   box each instead.\n2. **Track.** `picsart_media_track_motion` with the video's `source`, the window `t0`/`t1` in video\n   seconds, and `targets`: each one `bbox` (`[x, y, w, h]` at `t0`, in video pixels), `points`, or\n   `auto`. Set `tolerance` for the source (below).\n3. **Read the result before you trust it.** Per target it returns `animations` (the fitted\n   keyframes with their easing, or an expression), `static` (the channels that did not move), `meta` with each\n   channel's `maxError`, and `notes` on anything that weakened the track. `maxError` is the honest\n   reconstruction error: a fit with a large one is an approximation, and you say so.\n4. **Place it on a layer** (below), as keyframes with `timeMode: \"layer\"`.\n5. **Check it.** Export your render of the same window, track it with the same targets, and compare\n   the two with `picsart_media_track_motion_diff` (below).\n\n## What the tracker cannot see\n\nIt measures where texture goes, frame to frame. Some changes have no texture to follow:\n\n| Change | What to do instead |\n|---|---|\n| a reveal: a wipe, a mask growing | cropped stills of the box, one per frame of the reveal, for its edge; track whatever drives the reveal |\n| a fade or a twinkle | stills either side and through it; the tracker has no alpha (below) |\n| non-rigid motion: a flap, a bend, a wobble | track the centre only, and build the shape change as a loop |\n| a small idle loop: bobbing, swaying, breathing | track over several cycles, at least one and a half: a sway in rotation comes back as a `sine` expression; build any other loop from the cycles you measured |\n| motion blur on a fast move | track the rests either side; the smeared frames between them are not measurements |\n\n## Illustrated templates\n\nStickers, doodles and cut-out shapes over a photo or a flat background:\n\n- **A sticker with a white border tracks best with `\"estimator\": \"segmentation\"`**: the border is a\n  clean silhouette, where its flat fill gives features little to hold.\n\n## Tolerance, per channel and per unit\n\n`tolerance` is position in **pixels**, `toleranceScale` a **ratio** (default 0.01) and\n`toleranceRotation` **degrees** (default 0.5). Never reuse a pixel tolerance for scale: 1.5\n\"pixels\" once declared a 57% shrink static.\n\n| Source | `tolerance` |\n|---|---|\n| a compressed reference video | about 1.5 (it measures to about 0.5–2 px) |\n| your own render | 0.5 or less |\n\nToo tight on a compressed source and the keyframes chase codec noise.\n\n## Choosing targets\n\n- **Start where the object is solid.** During a fade-in its features are weak, the track dies on\n  the first frame and freezes. Start `t0` once it is fully opaque.\n- **A tight box inside the object's face.** A box over a tilted card in a stack catches its\n  neighbours, and the fit mixes two motions.\n- **Opaque objects only, for rotation and scale.** A half-transparent object blends with whatever\n  moves behind it, and its measured pose drifts toward that. Ghosts, depth-faded cards and\n  translucent chrome give approximate motion at best; say so.\n- **Video playing inside the box**: the features follow the content, not the box. Set `\"estimator\":\n  \"segmentation\"` on that target, which tracks the silhouette. Its centre leans toward bright\n  content, so a still element of the same group, tracked instead, is more exact.\n- **A rotating object takes a box, never points.** A point tracks translation only: a corner\n  turning in place reads about 4 px wrong at 30°. A box in `similarity` mode compensates.\n- **End the window before the object is covered.** When it slides behind something, its confidence\n  falls and its pose freezes. Shorten `t1` and track again, rather than keep the frozen tail.\n- **Coming in from off screen**: while the object is clipped by the frame edge, its centre and scale\n  are biased. Start once it is fully on screen.\n- **Clip the window, not the keyframes.** A fit over frozen samples emits false keyframes at the\n  tail. Track again with an earlier `t1`; never prune the keyframes by hand.\n\n| Content | `mode` | `estimator` |\n|---|---|---|\n| textured (a photo, a UI card with text) | `similarity` | `features` |\n| flat or solid, or video inside the box | `similarity` | `segmentation` |\n| stretched in perspective | `corners` | `features` |\n| position alone matters | `center` | either |\n\n## Reading the result\n\n- **Position is the object's centre**, in the video's own pixels. Under a layer's default anchor\n  (the centre of its box, for every content kind but text), that is the layer's\n  `transform.position`. Text anchors at its origin: place it by its alignment instead.\n- **Rotation and scale are relative to `t0`**: 0° and 1.0 there. The tracker cannot know an\n  object's tilt at `t0` from one box: add the pose the layer starts at yourself.\n- **Rotation is counter-clockwise positive; a scene's rotation is clockwise.** A turn the tracker\n  reports as +12° is `-12` in the scene. Flip the sign of every rotation value you place.\n- **Scale comes as `[s, s]`**, because a scene's scale is 2-D.\n- **Keyframe times count from `t0`** unless the call set `normalizeTime: \"absolute\"`.\n- **Easing is a preset or a fitted curve**, chosen per segment. The closest of the four presets\n  (`linear`, `ease_in`, `ease_out`, `ease_in_out`) comes first. When none of them fits within\n  `tolerance`, the segment moves and it has enough samples, it gets a fitted curve instead:\n  `{\"kind\": \"cubic_bezier\", \"control\": [x1, y1, x2, y2]}`, the scene's own parametric easing, with\n  a y outside 0–1 for an overshoot. Copy either form into the keyframe as it came. A short segment\n  keeps its closest preset, so a sharp snap or a long settle there reads close, not exact: its\n  `maxError` says how close; track it denser to give it a curve.\n- **An expression** instead of keyframes means the channel fit one within tolerance: `orbit` (the\n  path is a circle) or `sine` (a rotation that sways back and forth, `meta`'s `kind` says which). On\n  a compressed reference real paths often are not pure circles, and the keyframes it falls back to\n  are the honest answer, not a failure.\n- **The tracker has no alpha.** It cannot tell a fade from a move. Judge a fade on stills either side\n  of it, and never invent one to cover a jump in pose.\n\n## Placing tracked motion on a layer\n\nThe result is in the reference's frame and time. A layer needs its own:\n\n- **Time.** A keyframe's layer time is its tracked time, plus `t0` when times count from `t0`, plus\n  the scene second where the reference's second 0 plays, minus the layer's `start`.\n- **Pixels.** When the scene is not the reference's size, multiply each position by the scene's size\n  over the reference's (the same factor on both axes) and add the offset of where the reference's\n  frame sits in the scene.\n- **Pose.** Rotation: the layer's starting rotation minus the tracked value. Scale: the layer's\n  starting scale times the tracked value.\n- **Easing**: keep each keyframe's easing where it is; it shapes the segment that starts at it.\n\nAn expression is in the reference's time, and an orbit also in its pixels, so these rules cannot\nmove it: when the scene differs from the reference in size or timing, track that object again as\nkeyframes, or place it at the reference's size and time inside a composition sized like the\nreference.\n\n## Checking a render against the reference\n\nExport your render over the fragment, track it with the same targets and the same `mode`, and pass\nboth tracking results to `picsart_media_track_motion_diff` with the `target` and the `channel`.\nIt returns `maxDev`, `meanDev` and `worstT`, the time where the two differ most. A number replaces\n\"it looks about right\": tune until `maxDev` clears the threshold the use case sets.\n\n- **Like for like.** The same targets, the same window, the same tolerance on both sides. Your\n  render tracked from its own second 0 against the reference from its second 4 is a 4 s offset, not\n  a mismatch.\n- **What the deviation says.** A constant offset is geometry: a position or a size. A deviation\n  that grows and shrinks along a move is easing. One that peaks at a beat's start or end is timing.\n"
AFTER
"---\nname: motion-tracking-from-video\ndescription: Measure what objects visibly do in a video - where each one moves, turns and grows, frame by frame - with picsart_media_track_motion, and turn the measurement into scene keyframes; then measure your own render the same way and compare the two with picsart_media_track_motion_diff. Covers choosing targets and estimators, tolerances, reading the result honestly, and placing tracked motion on a layer. Use when a scene must reproduce motion seen in a reference video, or when a render's motion must be checked against one.\nmetadata:\n  ported-from: motion-extraction unversioned at 55.75.0\n---\n\n# Motion extraction\n\n`picsart_media_track_motion` recovers what objects **visibly do** in a flattened video: their\nposition, rotation and scale on every frame, fitted into scene keyframes. It does not recover the\nkeyframes someone authored. It produces new motion data that reproduces what is on screen.\n\nLooking decides **what** to track. The tracker measures **where**, every frame, to a fraction of a\npixel. Never write a motion's numbers from a still or from memory: every timing and distance comes\nfrom a tracking result.\n\n## Cost and receipts\n\nNeither `picsart_media_track_motion` nor `picsart_media_track_motion_diff` is a SPEND unless its\ndescription says it charges (picsart-video-rules), so track as often as a beat needs. Each replies\nwith a job receipt by default: follow it as picsart-video-rules says. Targets that share a tracking window go in one call, as many as\nits `targets` schema allows: one `t0`/`t1` holds for every target, and each must be fully visible at\n`t0`. A target that enters later, or is covered before `t1`, gets a call with its own window.\n\n## The method\n\n1. **Look at the first frame of the window** with `picsart_media_video_stills`, and choose what to\n   track. On a flat background, `{ \"auto\": true, \"max\": 9 }` finds the distinct objects at `t0` and\n   tracks each, ranked by how much they move. Objects that overlap merge into one, so give those a\n   box each instead.\n2. **Track.** `picsart_media_track_motion` with the video's `source`, the window `t0`/`t1` in video\n   seconds, and `targets`: each one `bbox` (`[x, y, w, h]` at `t0`, in video pixels), `points`, or\n   `auto`. Set `tolerance` for the source (below).\n3. **Read the result before you trust it.** Per target it returns `animations` (the fitted\n   keyframes with their easing, or an expression), `static` (the channels that did not move), `meta` with each\n   channel's `maxError`, and `notes` on anything that weakened the track. `maxError` is the honest\n   reconstruction error: a fit with a large one is an approximation, and you say so.\n4. **Place it on a layer** (below), as keyframes with `timeMode: \"layer\"`.\n5. **Check it.** Export your render of the same window, track it with the same targets, and compare\n   the two with `picsart_media_track_motion_diff` (below).\n\n## What the tracker cannot see\n\nIt measures where texture goes, frame to frame. Some changes have no texture to follow:\n\n| Change | What to do instead |\n|---|---|\n| a reveal: a wipe, a mask growing | cropped stills of the box, one per frame of the reveal, for its edge; track whatever drives the reveal |\n| a fade or a twinkle | stills either side and through it; the tracker has no alpha (below) |\n| non-rigid motion: a flap, a bend, a wobble | track the centre only, and build the shape change as a loop |\n| a small idle loop: bobbing, swaying, breathing | track over several cycles, at least one and a half: a sway in rotation comes back as a `sine` expression; build any other loop from the cycles you measured |\n| motion blur on a fast move | track the rests either side; the smeared frames between them are not measurements |\n\n## Illustrated templates\n\nStickers, doodles and cut-out shapes over a photo or a flat background:\n\n- **A sticker with a white border tracks best with `\"estimator\": \"segmentation\"`**: the border is a\n  clean silhouette, where its flat fill gives features little to hold.\n\n## Tolerance, per channel and per unit\n\n`tolerance` is position in **pixels**, `toleranceScale` a **ratio** (default 0.01) and\n`toleranceRotation` **degrees** (default 0.5). Never reuse a pixel tolerance for scale: 1.5\n\"pixels\" once declared a 57% shrink static.\n\n| Source | `tolerance` |\n|---|---|\n| a compressed reference video | about 1.5 (it measures to about 0.5–2 px) |\n| your own render | 0.5 or less |\n\nToo tight on a compressed source and the keyframes chase codec noise.\n\n## Choosing targets\n\n- **Start where the object is solid.** During a fade-in its features are weak, the track dies on\n  the first frame and freezes. Start `t0` once it is fully opaque.\n- **A tight box inside the object's face.** A box over a tilted card in a stack catches its\n  neighbours, and the fit mixes two motions.\n- **Opaque objects only, for rotation and scale.** A half-transparent object blends with whatever\n  moves behind it, and its measured pose drifts toward that. Ghosts, depth-faded cards and\n  translucent chrome give approximate motion at best; say so.\n- **Video playing inside the box**: the features follow the content, not the box. Set `\"estimator\":\n  \"segmentation\"` on that target, which tracks the silhouette. Its centre leans toward bright\n  content, so a still element of the same group, tracked instead, is more exact.\n- **A rotating object takes a box, never points.** A point tracks translation only: a corner\n  turning in place reads about 4 px wrong at 30°. A box in `similarity` mode compensates.\n- **End the window before the object is covered.** When it slides behind something, its confidence\n  falls and its pose freezes. Shorten `t1` and track again, rather than keep the frozen tail.\n- **Coming in from off screen**: while the object is clipped by the frame edge, its centre and scale\n  are biased. Start once it is fully on screen.\n- **Clip the window, not the keyframes.** A fit over frozen samples emits false keyframes at the\n  tail. Track again with an earlier `t1`; never prune the keyframes by hand.\n- **A track that is covered, then reacquired.** After the confidence collapses, the later samples can\n  be a valid track of the object that replaced it (a card becoming a chip). Check them against a\n  still before discarding them.\n- **One object, two stretches that need different estimators** (a video inside a card, then siblings\n  touching it): track each stretch on its own, with one overlapping frame. The two estimators'\n  reference points differ by about 2 px: measure that offset on the overlap frame and subtract it\n  when you join them, and multiply the second stretch's scale by the overlap frame's scale.\n\n| Content | `mode` | `estimator` |\n|---|---|---|\n| textured (a photo, a UI card with text) | `similarity` | `features` |\n| flat or solid, or video inside the box | `similarity` | `segmentation` |\n| stretched in perspective | `corners` | `features` |\n| position alone matters | `center` | either |\n\n## Reading the result\n\n- **Position is the object's centre**, in the video's own pixels. Under a layer's default anchor\n  (the centre of its box, for every content kind but text), that is the layer's\n  `transform.position`. Text anchors at its origin: place it by its alignment instead.\n- **Rotation and scale are relative to `t0`**: 0° and 1.0 there. The tracker cannot know an\n  object's tilt at `t0` from one box: add the pose the layer starts at yourself.\n- **Rotation is counter-clockwise positive; a scene's rotation is clockwise.** A turn the tracker\n  reports as +12° is `-12` in the scene. Flip the sign of every rotation value you place.\n- **Scale comes as `[s, s]`**, because a scene's scale is 2-D.\n- **Keyframe times count from `t0`** unless the call set `normalizeTime: \"absolute\"`.\n- **Easing is a preset or a fitted curve**, chosen per segment. The closest of the four presets\n  (`linear`, `ease_in`, `ease_out`, `ease_in_out`) comes first. When none of them fits within\n  `tolerance`, the segment moves and it has enough samples, it gets a fitted curve instead:\n  `{\"kind\": \"cubic_bezier\", \"control\": [x1, y1, x2, y2]}`, the scene's own parametric easing, with\n  a y outside 0–1 for an overshoot. Copy either form into the keyframe as it came. A short segment\n  keeps its closest preset, so a sharp snap or a long settle there reads close, not exact: its\n  `maxError` says how close; track it denser to give it a curve.\n- **An expression** instead of keyframes means the channel fit one within tolerance: `orbit` (the\n  path is a circle) or `sine` (a rotation that sways back and forth, `meta`'s `kind` says which). On\n  a compressed reference real paths often are not pure circles, and the keyframes it falls back to\n  are the honest answer, not a failure.\n- **The tracker has no alpha.** It cannot tell a fade from a move. Judge a fade on stills either side\n  of it, and never invent one to cover a jump in pose.\n\n## Placing tracked motion on a layer\n\nThe result is in the reference's frame and time. A layer needs its own:\n\n- **Time.** A keyframe's layer time is its tracked time, plus `t0` when times count from `t0`, plus\n  the scene second where the reference's second 0 plays, minus the layer's `start`.\n- **Pixels.** When the scene is not the reference's size, multiply each position by the scene's size\n  over the reference's (the same factor on both axes) and add the offset of where the reference's\n  frame sits in the scene.\n- **Pose.** Rotation: the layer's starting rotation minus the tracked value. Scale: the layer's\n  starting scale times the tracked value.\n- **Easing**: keep each keyframe's easing where it is; it shapes the segment that starts at it.\n\nAn expression is in the reference's time, and an orbit also in its pixels, so these rules cannot\nmove it: when the scene differs from the reference in size or timing, track that object again as\nkeyframes, or place it at the reference's size and time inside a composition sized like the\nreference.\n\n## Checking a render against the reference\n\nExport your render over the fragment, track it with the same targets and the same `mode`, and pass\nboth tracking results to `picsart_media_track_motion_diff` with the `target` and the `channel`.\nIt returns `maxDev`, `meanDev` and `worstT`, the time where the two differ most. A number replaces\n\"it looks about right\": tune until `maxDev` clears the threshold the use case sets.\n\n- **Like for like.** The same targets, the same window, the same tolerance on both sides. Your\n  render tracked from its own second 0 against the reference from its second 4 is a 4 s offset, not\n  a mismatch.\n- **What the deviation says.** A constant offset is geometry: a position or a size. A deviation\n  that grows and shrinks along a move is easing. One that peaks at a beat's start or end is timing.\n\n## Measuring for a seam\n\nWhen a seam decision depends on a reference's velocity, path or overlap, the measurement feeds\n`motion-scene-continuity` (fetch with `picsart_scene_recipe`). Tracking answers what moved, where\nand when; continuity judges what that movement communicates and whether the scene should preserve,\nreplace, interrupt or reset it. A close numeric fit does not by itself prove a readable handoff.\n"
Full snapshot data
{
  "description": "Measure what objects visibly do in a video - where each one moves, turns and grows, frame by frame - with picsart_media_track_motion, and turn the measurement into scene keyframes; then measure your own render the same way and compare the two with picsart_media_track_motion_diff. Covers choosing targets and estimators, tolerances, reading the result honestly, and placing tracked motion on a layer. Use when a scene must reproduce motion seen in a reference video, or when a render's motion must be checked against one.",
  "included_files": [],
  "name": "motion-tracking-from-video",
  "skill_md_contents": "---\nname: motion-tracking-from-video\ndescription: Measure what objects visibly do in a video - where each one moves, turns and grows, frame by frame - with picsart_media_track_motion, and turn the measurement into scene keyframes; then measure your own render the same way and compare the two with picsart_media_track_motion_diff. Covers choosing targets and estimators, tolerances, reading the result honestly, and placing tracked motion on a layer. Use when a scene must reproduce motion seen in a reference video, or when a render's motion must be checked against one.\nmetadata:\n  ported-from: motion-extraction unversioned at 55.75.0\n---\n\n# Motion extraction\n\n`picsart_media_track_motion` recovers what objects **visibly do** in a flattened video: their\nposition, rotation and scale on every frame, fitted into scene keyframes. It does not recover the\nkeyframes someone authored. It produces new motion data that reproduces what is on screen.\n\nLooking decides **what** to track. The tracker measures **where**, every frame, to a fraction of a\npixel. Never write a motion's numbers from a still or from memory: every timing and distance comes\nfrom a tracking result.\n\n## Cost and receipts\n\nNeither `picsart_media_track_motion` nor `picsart_media_track_motion_diff` is a SPEND unless its\ndescription says it charges (picsart-video-rules), so track as often as a beat needs. Each replies\nwith a job receipt by default: follow it as picsart-video-rules says. Targets that share a tracking window go in one call, as many as\nits `targets` schema allows: one `t0`/`t1` holds for every target, and each must be fully visible at\n`t0`. A target that enters later, or is covered before `t1`, gets a call with its own window.\n\n## The method\n\n1. **Look at the first frame of the window** with `picsart_media_video_stills`, and choose what to\n   track. On a flat background, `{ \"auto\": true, \"max\": 9 }` finds the distinct objects at `t0` and\n   tracks each, ranked by how much they move. Objects that overlap merge into one, so give those a\n   box each instead.\n2. **Track.** `picsart_media_track_motion` with the video's `source`, the window `t0`/`t1` in video\n   seconds, and `targets`: each one `bbox` (`[x, y, w, h]` at `t0`, in video pixels), `points`, or\n   `auto`. Set `tolerance` for the source (below).\n3. **Read the result before you trust it.** Per target it returns `animations` (the fitted\n   keyframes with their easing, or an expression), `static` (the channels that did not move), `meta` with each\n   channel's `maxError`, and `notes` on anything that weakened the track. `maxError` is the honest\n   reconstruction error: a fit with a large one is an approximation, and you say so.\n4. **Place it on a layer** (below), as keyframes with `timeMode: \"layer\"`.\n5. **Check it.** Export your render of the same window, track it with the same targets, and compare\n   the two with `picsart_media_track_motion_diff` (below).\n\n## What the tracker cannot see\n\nIt measures where texture goes, frame to frame. Some changes have no texture to follow:\n\n| Change | What to do instead |\n|---|---|\n| a reveal: a wipe, a mask growing | cropped stills of the box, one per frame of the reveal, for its edge; track whatever drives the reveal |\n| a fade or a twinkle | stills either side and through it; the tracker has no alpha (below) |\n| non-rigid motion: a flap, a bend, a wobble | track the centre only, and build the shape change as a loop |\n| a small idle loop: bobbing, swaying, breathing | track over several cycles, at least one and a half: a sway in rotation comes back as a `sine` expression; build any other loop from the cycles you measured |\n| motion blur on a fast move | track the rests either side; the smeared frames between them are not measurements |\n\n## Illustrated templates\n\nStickers, doodles and cut-out shapes over a photo or a flat background:\n\n- **A sticker with a white border tracks best with `\"estimator\": \"segmentation\"`**: the border is a\n  clean silhouette, where its flat fill gives features little to hold.\n\n## Tolerance, per channel and per unit\n\n`tolerance` is position in **pixels**, `toleranceScale` a **ratio** (default 0.01) and\n`toleranceRotation` **degrees** (default 0.5). Never reuse a pixel tolerance for scale: 1.5\n\"pixels\" once declared a 57% shrink static.\n\n| Source | `tolerance` |\n|---|---|\n| a compressed reference video | about 1.5 (it measures to about 0.5–2 px) |\n| your own render | 0.5 or less |\n\nToo tight on a compressed source and the keyframes chase codec noise.\n\n## Choosing targets\n\n- **Start where the object is solid.** During a fade-in its features are weak, the track dies on\n  the first frame and freezes. Start `t0` once it is fully opaque.\n- **A tight box inside the object's face.** A box over a tilted card in a stack catches its\n  neighbours, and the fit mixes two motions.\n- **Opaque objects only, for rotation and scale.** A half-transparent object blends with whatever\n  moves behind it, and its measured pose drifts toward that. Ghosts, depth-faded cards and\n  translucent chrome give approximate motion at best; say so.\n- **Video playing inside the box**: the features follow the content, not the box. Set `\"estimator\":\n  \"segmentation\"` on that target, which tracks the silhouette. Its centre leans toward bright\n  content, so a still element of the same group, tracked instead, is more exact.\n- **A rotating object takes a box, never points.** A point tracks translation only: a corner\n  turning in place reads about 4 px wrong at 30°. A box in `similarity` mode compensates.\n- **End the window before the object is covered.** When it slides behind something, its confidence\n  falls and its pose freezes. Shorten `t1` and track again, rather than keep the frozen tail.\n- **Coming in from off screen**: while the object is clipped by the frame edge, its centre and scale\n  are biased. Start once it is fully on screen.\n- **Clip the window, not the keyframes.** A fit over frozen samples emits false keyframes at the\n  tail. Track again with an earlier `t1`; never prune the keyframes by hand.\n- **A track that is covered, then reacquired.** After the confidence collapses, the later samples can\n  be a valid track of the object that replaced it (a card becoming a chip). Check them against a\n  still before discarding them.\n- **One object, two stretches that need different estimators** (a video inside a card, then siblings\n  touching it): track each stretch on its own, with one overlapping frame. The two estimators'\n  reference points differ by about 2 px: measure that offset on the overlap frame and subtract it\n  when you join them, and multiply the second stretch's scale by the overlap frame's scale.\n\n| Content | `mode` | `estimator` |\n|---|---|---|\n| textured (a photo, a UI card with text) | `similarity` | `features` |\n| flat or solid, or video inside the box | `similarity` | `segmentation` |\n| stretched in perspective | `corners` | `features` |\n| position alone matters | `center` | either |\n\n## Reading the result\n\n- **Position is the object's centre**, in the video's own pixels. Under a layer's default anchor\n  (the centre of its box, for every content kind but text), that is the layer's\n  `transform.position`. Text anchors at its origin: place it by its alignment instead.\n- **Rotation and scale are relative to `t0`**: 0° and 1.0 there. The tracker cannot know an\n  object's tilt at `t0` from one box: add the pose the layer starts at yourself.\n- **Rotation is counter-clockwise positive; a scene's rotation is clockwise.** A turn the tracker\n  reports as +12° is `-12` in the scene. Flip the sign of every rotation value you place.\n- **Scale comes as `[s, s]`**, because a scene's scale is 2-D.\n- **Keyframe times count from `t0`** unless the call set `normalizeTime: \"absolute\"`.\n- **Easing is a preset or a fitted curve**, chosen per segment. The closest of the four presets\n  (`linear`, `ease_in`, `ease_out`, `ease_in_out`) comes first. When none of them fits within\n  `tolerance`, the segment moves and it has enough samples, it gets a fitted curve instead:\n  `{\"kind\": \"cubic_bezier\", \"control\": [x1, y1, x2, y2]}`, the scene's own parametric easing, with\n  a y outside 0–1 for an overshoot. Copy either form into the keyframe as it came. A short segment\n  keeps its closest preset, so a sharp snap or a long settle there reads close, not exact: its\n  `maxError` says how close; track it denser to give it a curve.\n- **An expression** instead of keyframes means the channel fit one within tolerance: `orbit` (the\n  path is a circle) or `sine` (a rotation that sways back and forth, `meta`'s `kind` says which). On\n  a compressed reference real paths often are not pure circles, and the keyframes it falls back to\n  are the honest answer, not a failure.\n- **The tracker has no alpha.** It cannot tell a fade from a move. Judge a fade on stills either side\n  of it, and never invent one to cover a jump in pose.\n\n## Placing tracked motion on a layer\n\nThe result is in the reference's frame and time. A layer needs its own:\n\n- **Time.** A keyframe's layer time is its tracked time, plus `t0` when times count from `t0`, plus\n  the scene second where the reference's second 0 plays, minus the layer's `start`.\n- **Pixels.** When the scene is not the reference's size, multiply each position by the scene's size\n  over the reference's (the same factor on both axes) and add the offset of where the reference's\n  frame sits in the scene.\n- **Pose.** Rotation: the layer's starting rotation minus the tracked value. Scale: the layer's\n  starting scale times the tracked value.\n- **Easing**: keep each keyframe's easing where it is; it shapes the segment that starts at it.\n\nAn expression is in the reference's time, and an orbit also in its pixels, so these rules cannot\nmove it: when the scene differs from the reference in size or timing, track that object again as\nkeyframes, or place it at the reference's size and time inside a composition sized like the\nreference.\n\n## Checking a render against the reference\n\nExport your render over the fragment, track it with the same targets and the same `mode`, and pass\nboth tracking results to `picsart_media_track_motion_diff` with the `target` and the `channel`.\nIt returns `maxDev`, `meanDev` and `worstT`, the time where the two differ most. A number replaces\n\"it looks about right\": tune until `maxDev` clears the threshold the use case sets.\n\n- **Like for like.** The same targets, the same window, the same tolerance on both sides. Your\n  render tracked from its own second 0 against the reference from its second 4 is a 4 s offset, not\n  a mismatch.\n- **What the deviation says.** A constant offset is geometry: a position or a size. A deviation\n  that grows and shrinks along a move is easing. One that peaks at a beat's start or end is timing.\n\n## Measuring for a seam\n\nWhen a seam decision depends on a reference's velocity, path or overlap, the measurement feeds\n`motion-scene-continuity` (fetch with `picsart_scene_recipe`). Tracking answers what moved, where\nand when; continuity judges what that movement communicates and whether the scene should preserve,\nreplace, interrupt or reset it. A close numeric fit does not by itself prove a readable handoff.\n"
}

SHA-256 of public snapshot: 922786c5d4e31b95667ae333ab6ba1ad464917cabdb7069c46b36f3f7e4bd0c8