101 Camera Moves

Every move named, with the words that ask for it — and the ones you should never pay to generate.

101moves
101 / 101written
4 / 101verified
12free in post
This is free, and here is the honest reason. aicameramovements.com already gives away 46 of these with looping video previews, in this same four-part form. If you want to SEE a move before you write it, go there — it is better than this page at that job, and it costs nothing.

What it cannot tell you is which of these you should never pay to generate. Its revenue comes from the "turn into video" button, so "do this locally for nothing" is the one sentence it can never print. 12 of these 101 moves are edit decisions, not generation problems — the ffmpeg is on the card, and generating them costs you a re-roll for a worse result.

Every phrasing here was written by us from the four-part form. 4 of 101 have actually been run and measured — that number is printed rather than buried, and each finding names the route we measured it on, because your platform may gate things differently from ours.
prompt
The model can do this from words alone. Ask for it. Costs a generation.
free in post
Do not generate this. It is an edit, not a generation — a keyframe in CapCut, Premiere, DaVinci or any editor. Paying a model for it costs you a re-roll and lands worse. The ffmpeg on the card is for people who prefer a command line; you do not need one.
hybrid
Generate the shot, then finish the move in your editor. Half of it costs nothing.
risky
The model can do this but often does not. Budget re-rolls, or route it elsewhere. The card says what usually goes wrong.
verified
We ran this one and measured the result, on a named provider. Everything without the mark is written but not yet run — the counter above says how many.

All 101 moves

001Panswivel left/right on a fixed spot.promptverified
Use it when
Use to carry the eye across a scene from one vantage point when nothing needs to separate in depth.
The phrasing
Pan. Movement: the camera holds one position on the ground and rotates on its vertical axis, sweeping the frame left or right across the scene; every object keeps its own place in the world and none of them travel toward the lens. Speed: even rate, about 30 degrees of rotation across 4 seconds, with no change of pace at either end. Framing: the horizon stays parallel to the frame edges and the subject stays inside the frame for the whole sweep. End: settles with the subject on the third opposite the direction of travel and open space ahead of it.
Watch for. A pan is a rotation, so it produces almost no parallax: measured 1.07:1 near-to-far on a plate whose depth map read as correct. If foreground and background need to separate, truck instead of panning.
Measured. A pan is a ROTATION and separates nothing at any depth. Measured 1.07:1 on a plate whose depth map looked perfect. Never use a pan when depth is the point.
002Tiltnod up/down on a fixed spot.prompt
Use it when
Use to move between two heights of the same scene, such as ground to face or face to sky.
The phrasing
Tilt. Movement: the camera holds one position and rotates on its horizontal axis, the frame nodding up or down through the scene; objects hold their own positions and the ground line crosses the frame once. Speed: one continuous nod of about 25 degrees over 3 seconds, easing over the last half second. Framing: verticals stay parallel to the frame edges and the subject stays inside the frame for the whole nod. End: settles on whatever the rotation arrives at, that element centred with about a third of the frame above it.
Watch for. Untested on our stack. Like a pan it is a rotation and separates nothing in depth, and on a tall subject the far end of the nod keystones the verticals it was meant to keep straight.
003Dolly incamera physically forward toward the subject.promptverified
Use it when
Use to commit attention to one subject over a beat without cutting.
The phrasing
Dolly in. Movement: the camera travels forward along its own axis toward the subject; the subject keeps its place in the world and grows in the frame because the camera closed the distance, not because it moved at the lens. Speed: steady walking pace, closing about half the starting distance over 5 seconds. Framing: the eyeline stays on the same third of the frame the whole way and the feet leave the frame before the chest does. End: finishes on a chest-up framing with the subject centred.
Watch for. Without the causality clause the model flies the object at the lens instead of moving the camera, measured 20 Sep 2026 at $1.84. Keep the words about the camera closing the distance in every version of the prompt.
Measured. MEASURED 20 Sep 2026 ($1.84): renders as specified when the prompt carries the CAUSALITY clause -- the subject grows because the lens approached it. Without it the object flies at the lens (rule 32).
004Dolly outcamera physically backward away.prompt
Use it when
Use to give a subject its context, or to close a beat by opening the space around it.
The phrasing
Dolly out. Movement: the camera travels backward along its own axis away from the subject; the subject keeps its place in the world and more of the room enters the frame around it as the distance opens. Speed: steady, about 2 metres over 5 seconds, no stop and start. Framing: the subject stays on the same vertical axis of the frame and its head stays legible as it gets smaller. End: finishes wide with the subject about a third of the frame height and both floor and back wall in shot.
Watch for. Untested on our stack. Backing out asks the model for geometry it never committed to, so the walls and floor entering the frame often re-invent themselves mid-move; keep the retreat short or establish the wide first.
005Truck leftcamera parallel left through space.promptverified
Use it when
Use when foreground and background must visibly separate, which is the one thing a rotation cannot do.
The phrasing
Truck left. Movement: the camera keeps its facing and slides bodily to the left, parallel to the subject plane; near objects cross the frame faster than far ones and every object holds its own place in the world. Speed: steady sidestep, about 1.5 metres over 4 seconds, one continuous slide. Framing: the subject stays at the same height in the frame and the near edge object stays a defined shape rather than a smear. End: finishes with the subject on the right third and the space the camera moved into filling the left of the frame.
Watch for. This is the move that produces real parallax and the reason to reject a pan: panned client plates measured 1.10-1.23 near-to-far against 4.5-9.1 for generated travel. When briefing a shoot, say walk past the subject rather than stand and turn.
Measured. MEASURED 20 Sep 2026 (parallax.py): a TRUCK is what produces real parallax. Our panned client plates measured 1.10-1.23 near:far; generated dollies 4.5-9.1. Shooting note for any client: WALK past the subject, do not stand and pan.
006Truck rightcamera parallel right.promptverified
Use it when
Use as the mirror of truck left when the reveal or the eye trace runs the other way.
The phrasing
Truck right. Movement: the camera keeps its facing and slides bodily to the right, parallel to the subject plane; near objects cross the frame faster than far ones and every object holds its own place in the world. Speed: steady sidestep, about 1.5 metres over 4 seconds, one continuous slide. Framing: the subject stays at the same height in the frame and the near edge object stays a defined shape rather than a smear. End: finishes with the subject on the left third and the space the camera moved into filling the right of the frame.
Watch for. Same measurement as truck left: it separates depth where a pan does not. Pick the direction from what enters the frame, since the trailing edge is where the previous shot's subject appears to exit.
Measured. See truck left -- same measurement.
007Pedestal upcamera straight up, no tilt.prompt
Use it when
Use to change the vantage on a subject without changing the angle the lens looks at it.
The phrasing
Pedestal up. Movement: the camera rises vertically on its column, the lens keeping the same angle through the whole climb so the frame gains height without nodding; the subject holds its place and is seen from further above as the camera climbs. Speed: rises about 1.2 metres over 4 seconds at an even rate. Framing: verticals stay parallel to the frame edges and the subject stays in the frame through the rise. End: finishes looking across the subject from above eye height, with the ground line lower in the frame than it started.
Watch for. Untested on our stack. Models substitute a tilt for a rise, so the frame ends up angled down from the old height instead of level from a new one; say the lens angle is carried unchanged and check the verticals at the last frame.
008Pedestal downcamera straight down, no tilt.prompt
Use it when
Use to arrive at a low vantage on a subject or a product without tipping the lens.
The phrasing
Pedestal down. Movement: the camera descends vertically on its column, the lens keeping the same angle through the whole drop so the frame loses height without nodding; the subject holds its place and is seen from further below as the camera drops. Speed: descends about 1.2 metres over 4 seconds at an even rate. Framing: verticals stay parallel to the frame edges and the subject stays in the frame through the drop. End: finishes looking across the subject from below eye height, with more of the ground plane across the bottom of the frame.
Watch for. Untested on our stack. The same substitution as pedestal up: the model tilts down from the old height instead of dropping to a new one, and the floor arrives keystoned.
009Whip pan (flash pan)pan so fast the background blurs; an in-camera transitionprompt
Use it when
Use as an in-camera transition between two setups when a hard cut would read as too clean.
The phrasing
Whip pan (flash pan). Movement: the camera rotates on its spot fast enough that the scene smears into horizontal streaks, carrying the frame from one subject to another; the world holds its positions and only the rotation rate makes the blur. Speed: about 90 degrees in under half a second, accelerating out of the opening frames and decelerating into the last three. Framing: the opening subject stays readable for the first three frames and the arriving subject is readable within three frames of the smear clearing. End: finishes on a new composition with the arriving subject centred and no streaking left in the frame.
Watch for. Untested on our stack. Providers under-deliver the smear and give a fast pan instead, so state the streaking as the visible result rather than asking for speed. If the whip hides a cut, both halves must be generated at the same lens height or the join shows.
010Whip tiltviolent up/down whip to reveal a new scene or skyprompt
Use it when
Use to jump between two heights of a scene, most often ground to sky, as a transition rather than a cut.
The phrasing
Whip tilt. Movement: the camera rotates on its horizontal axis fast enough that the scene smears into vertical streaks, carrying the frame from one height to another and uncovering what is above or below; the world holds its positions through the rotation. Speed: about 70 degrees in under half a second, accelerating out of the opening frames and decelerating into the last three. Framing: the departing subject is readable for the first three frames and the arriving one within three frames of the smear clearing. End: finishes on the new height with the arriving subject centred and the horizon parallel to the frame edges.
Watch for. Untested on our stack. The vertical smear is harder for a model than the horizontal one and often comes back as a plain fast tilt; name the streaks. Watch the horizon at the landing, which tends to arrive off level.
011Snap zoominstant zoom onto a face for emphasisprompt
Use it when
Use to land a punctuation mark on a reaction or a line, usually cut to a sound.
The phrasing
Snap zoom. Movement: the field of view collapses onto the face in one step through the lens, so the framing tightens while the perspective and the distances between objects stay as they were; the subject keeps its place in the world. Speed: from the wide size to the close size in about three frames, and the rest of the beat runs at the close size. Framing: the face stays on the same point of the frame across the step and the eyeline does not jump between the two sizes. End: finishes on a head-and-shoulders framing with the eyes on the upper third.
Watch for. Untested on our stack. Models render a zoom as a push-in, so the perspective shifts and it reads as travel; if the body must not travel, route the beat to seedance 1.x cameraFixed and do the size change as a crop-and-scale locally, which is free and exact.
012Ping-pong panrapid pan between two talking headsrisky
Use it when
Use for a two-hander where the exchange between the pair is the content and a cut would slow it.
The phrasing
Ping-pong pan. Movement: the camera rotates on its spot between two people standing apart, swinging the frame from one to the other and back again; both people hold their positions and only the camera crosses the gap. Speed: each swing crosses in about 0.3 seconds, and between swings the camera drifts a few degrees on the face it landed on rather than arriving and parking. Framing: both faces are legible in the frames that land on them and both eyelines sit at the same height. End: finishes landed on one of the two in a chest-up framing, with the other's shoulder on the near edge of the frame.
Watch for. Identity drift with two people: the second face borrows features from the first every time the camera returns to it. Lock each person with their own reference plate, exactly one face per plate, count the people in the prompt, and compare the last landing against the first before accepting the take.
From our notes
needs two stable identities
013Party panerratic loose handheld matched to musicprompt
Use it when
Use for crowd, event and music-led material where the camera is a participant rather than an observer.
The phrasing
Party pan. Movement: the camera is hand-carried and swings loosely left and right off a drifting centre, the body shifting a few inches with each change of direction; the scene holds its own positions while the frame wanders across it. Speed: about two changes of direction a second, 15 to 25 degrees each way, and the amounts deliberately unequal. Framing: the subject stays within the middle half of the frame through the wander and faces stay legible while the frame is moving. End: finishes off axis with the subject about a third in from one edge and the frame line a few degrees off level.
Watch for. Under-specified handheld defaults to near-stillness, so the count of direction changes and the degrees are the whole instruction. The word erratic on its own buys nothing, and the wander tends to drift the subject out of the frame by the last second.
014Crash push-inrunning straight at the subject on a gimbalprompt
Use it when
Use to open on force, or to arrive on a subject at the top of a beat.
The phrasing
Crash push-in. Movement: the camera travels forward at running pace on a carried rig, the subject holding its ground while the distance closes; the frame bobs a little with each step and the subject grows because the camera arrived, not because it moved at the lens. Speed: closes about 4 metres in 1.5 seconds, accelerating through the first third and carrying that rate to the end. Framing: the face stays in the upper middle of the frame through the bob and the horizon stays within 5 degrees of level. End: finishes on a head-and-shoulders framing, face centred, about an arm's length out.
Watch for. Untested on our stack. Same causality failure as the dolly, worse at speed: state that the camera runs at the subject or the subject is pulled at the lens. Fast forward travel also warps the background, so keep the background simple or short.
015Slingshotslow push, then a fast snap to close-uphybrid
Use it when
Use when a beat has to start patient and arrive suddenly, usually cut to a drop.
The phrasing
Slingshot. Movement: the camera travels forward along one straight line toward the subject and keeps travelling to a close framing, with no change of direction and no change of rate; the subject holds its place and the distance closes ahead of the lens. Speed: generate the whole travel at one even rate over 6 seconds, since the slow-then-snap is applied afterwards. Framing: the face stays on the same third of the frame for the whole travel and stays legible at every size it passes through. End: finishes on a face-filling framing with the eyes on the upper third.
Watch for. The ramp belongs in post: generate one even-rate travel and do the retime locally, because a ramp asked for in the prompt lands on the wrong second and cannot be moved afterwards. Generate 20 to 30 percent more travel than the final needs so the retime has frames to spend.
From our notes
the ramp is POST
016Pendulum swingarc back and forth over the subjectprompt
Use it when
Use to hold on one subject for a long beat while the frame keeps changing.
The phrasing
Pendulum swing. Movement: the camera swings along an arc that passes over the subject and out the other side, then returns along the same arc; the subject holds its place at the centre of the arc and the camera carries the whole movement. Speed: one pass across in 2 seconds and the return in 2 seconds, faster through the middle of the arc than at either end. Framing: the subject stays at the centre of the frame through both passes and the ground line rises and falls behind it. End: finishes at one end of the arc looking down at the subject from about 45 degrees.
Watch for. Untested on our stack. An arc over a subject comes back as the subject rotating instead of the camera travelling, so state that the feet stay planted and the camera carries the arc. The overhead part of the pass loses the face, so put no dialogue there.
017Quick jolta brief shake on the beat dropfree in post
Use it when
Use to mark a beat drop or an impact already present in the audio.
What it should look like
Quick jolt. Movement: the whole frame displaces a few pixels in one direction and returns, so the image shifts as a unit while nothing in the world changes position. Speed: reaches a displacement of 8 to 12 pixels in one frame and returns over the next three. Framing: type and faces stay inside the safe area at maximum displacement and no frame edge is exposed. End: finishes back on the original framing with nothing cropped.
Do this in your editor — it costs nothing

A keyframe in CapCut, Premiere, DaVinci or any editor does this. The command below is the same move for anyone who prefers a terminal — you do not need one.

ffmpeg -y -i in.mp4 -vf "scale=1188:2112:flags=lanczos,crop=1080:1920:x='(in_w-1080)/2+if(between(t,2.000,2.180),26*sin(2*PI*36*(t-2.000))*(1-(t-2.000)/0.180),0)':y='(in_h-1920)/2+if(between(t,2.000,2.180),20*sin(2*PI*29*(t-2.000)+1.1)*(1-(t-2.000)/0.180),0)',setsar=1" -c:v libx264 -crf 17 -preset medium -pix_fmt yuv420p -c:a copy out.mp4
What to change
2.000 is the hit. It appears six times and every one must change together. Get it frame-accurate by computing it as frame_number/fps — at 30fps, frame 61 is 2.0333, not 2.03. 0.180 is how long the jolt takes to die; it appears four times. 26 and 20 are the kick amplitude in pixels, horizontal and vertical — 26 is a firm hit on a 1080-wide frame, 40+ is a slam, 12 is a nudge. 36 and 29 are the shake frequencies in Hz; keeping them different and non-integer stops the x and y wobbles locking into a clean circle, which reads as fake. 1188:2112 is the source scaled up 10% to give the crop somewhere to move — the amplitude can never exceed (1188-1080)/2 = 54px, so raise the scale if you raise the amplitude. Better: drop the scale entirely and feed it a 2160x3840 source, because this upscale softens the whole clip to buy 54 pixels.
Why not generate it
This has to be frame-accurate against the audio, and a generation cannot be aimed at a frame — the move's own ledger note says exactly that. Paying for a shake also means paying for the model to re-render every other pixel in the shot, so it will change things you had already approved.
Not run by us. This command is written from the documented filters, not executed here. Check it on a test clip before a client job.
Watch for. Do not generate this. It is a local post move, an offset of the finished frame, and it has to land on an exact audio frame, which no provider can hit; generating it also burns a take. Scale the shot 2 to 4 percent first so the offset never exposes an edge.
From our notes
must be frame-accurate to audio
018Ground-up revealstart at the feet, rise to the faceprompt
Use it when
Use as a character opener where the face is the payoff of the shot rather than its start.
The phrasing
Ground-up reveal. Movement: the camera starts low at the subject's feet and climbs, the lens angling up as it rises, so the body passes through the frame from shoes to knees to hands to chest to face; the subject holds its stance and the camera carries the climb. Speed: covers the body in 3 seconds, easing over the last half metre. Framing: the body stays centred across the frame through the climb and the face is legible for the final second. End: finishes on a low-angle chest-up framing, the camera below eye height and looking slightly up.
Watch for. Untested on our stack. An under-specified rise defaults to near-stillness or breaks into a jump, so name the body parts it passes in order and state it as one continuous climb.
019Sky-down revealstart on sky, tilt down as they enterprompt
Use it when
Use to open a scene on place before it opens on a person.
The phrasing
Sky-down reveal. Movement: the camera starts framed on open sky and rotates down through the horizon as the subject walks into the lower part of the frame; the sky and the ground hold their positions and the subject covers its own ground on foot. Speed: about 60 degrees of rotation over 3 seconds, with the subject entering the frame during the final second of it. Framing: the horizon crosses the frame once and stays level, and the subject is legible from the frame it enters. End: finishes on a chest-up framing at eye height with the subject centred and a strip of sky along the top edge.
Watch for. Untested on our stack. The arrival has to be possible within the rotation: a timecode places an event but not an impossible one, so write the rotation and the walk on the same timeline rather than asking for the subject to already be there.
020Model revealcurve around one person to reveal a secondrisky
Use it when
Use when the second person is the point of the shot and a cut would give them away early.
The phrasing
Model reveal. Movement: the camera curves around the first person on a short arc, keeping them in the frame while the arc opens the space behind them and brings a second person into view; both people hold their positions and the camera carries the curve. Speed: about 90 degrees of arc over 5 seconds at an even rate. Framing: the first face stays legible until the second enters and both are in the frame together for the final 2 seconds. End: finishes on a two-shot, the first person in profile on the near edge and the second facing the lens.
Watch for. Two identities in one take drift, and the second face borrows from the first. Give each person their own reference plate with exactly one face on it, count the people explicitly in the prompt or the model returns a duplicate of the first, and check both faces against their plates at the last frame.
From our notes
two identities
021Object pass-bytrack past a wall or pole to reveal the subjectprompt
Use it when
Use to reveal a subject while proving the depth of the space it stands in.
The phrasing
Object pass-by. Movement: the camera slides sideways so a near object such as a pole, a column or a wall edge crosses the frame and clears it, uncovering the subject standing behind; the near object and the subject both hold their places and only the camera travels. Speed: steady sidestep, the near object crossing the frame over about 1.5 seconds. Framing: the near object stays a defined silhouette rather than a smear as it crosses, and the subject is legible within three frames of it clearing. End: finishes with the subject centred and the near object gone off the trailing edge.
Watch for. Keep it a sideways slide, not a rotation: a pan uncovers nothing, measured 1.07:1 near-to-far, while sideways travel is what separates the planes. If the near object smears, it is travelling too fast to read as foreground.
From our notes
strong for DEPTH
022Mirror slippan past a reflection, tilt down to the real subjectrisky
Use it when
Use to show a subject twice in one take, the reflection first and the person second.
The phrasing
Mirror slip. Movement: the camera rotates across a reflective surface holding the subject's reflection, then tilts down and across to the subject standing in front of it; the person and the surface hold their places and the camera carries both movements as one. Speed: the sweep across the reflection in 2 seconds, then the tilt down in 1 second, run without a break between them. Framing: the reflected face is legible through the first half and the real face through the final second, with the edge of the reflective surface visible at the handover. End: finishes on a chest-up framing of the person with the edge of the glass along the top of the frame.
Watch for. Reflections drift: the reflected face stops matching the real one and the geometry of which way it faces breaks as soon as the camera moves. Generate short, compare the two faces frame by frame, and be ready to build the reflection as a separate element rather than asking one take for both.
From our notes
reflections drift
023Over-the-shoulder swivelfrom behind the back, rotate to the faceprompt
Use it when
Use to arrive on a face from behind, where the reveal is the identity of the person already on screen.
The phrasing
Over-the-shoulder swivel. Movement: the camera starts behind the subject's shoulder and arcs around the body to the front; the subject keeps their feet planted and their stance unchanged while the camera carries the whole rotation around them. Speed: about 150 degrees over 4 seconds at an even rate, easing through the final 30 degrees. Framing: the back of the head and the shoulder line stay in the frame through the arc and the face is legible for the final second. End: finishes on a front three-quarter chest-up framing, the shoulder that started nearest now on the frame edge.
Watch for. Untested on our stack. Models turn the person instead of travelling the camera, so state that the feet stay planted. The face only arrives at the end, so lock the identity with a plate or the payoff frame is a stranger.
024Lens block dropa hand blocks the lens, then pulls awayprompt
Use it when
Use as an in-camera transition that also gives a free cut point in the covered frames.
The phrasing
Lens block drop. Movement: one hand enters from a frame edge and covers the lens until the palm fills the frame, then withdraws the way it came and uncovers the scene behind it; the scene holds its positions throughout and only the hand and the camera move. Speed: the hand covers the frame in about four frames, the palm fills it for about six, and it clears in four. Framing: the palm fills the frame corner to corner at the covered point with no gap, and the scene behind is legible within three frames of the clear. End: finishes on the uncovered scene with the subject centred.
Watch for. Untested on our stack. Count the hands explicitly, since every duplicate-object failure is a missing count, and check the corners at the covered frame because a gap there kills the cut point the move exists for.
From our notes
also a free cut point
025Horizon flipupside down, roll 180° into normalrisky
Use it when
Use to open disoriented and resolve into a normal frame within one shot.
The phrasing
Horizon flip. Movement: the camera starts inverted and rolls 180 degrees on its lens axis until the horizon is level; the world keeps its own orientation and gravity, and the camera carries the entire roll. Speed: 180 degrees over 2.5 seconds, easing through the final 40 degrees. Framing: the subject stays at the centre of the frame through the roll and the horizon is a readable line at both ends. End: finishes level, horizon parallel to the frame edges, subject centred.
Watch for. Models roll the world instead of the camera: gravity inverts, hair and clothing fall upward and loose objects drift. Generate level and roll in post, which is free and frame-exact, unless something in the shot has to react to the inversion.
026Door close pushmove through a doorway as it opens or closesprompt
Use it when
Use to move between two spaces in one take, or to close a beat as the door takes the frame.
The phrasing
Door close push. Movement: the camera travels forward through a doorway while the door swings, the door frame passing the lens on both sides as the camera crosses the threshold and the room beyond opens around it; the door, the frame and the room all hold their own places. Speed: crosses the threshold in about 2 seconds at an even rate, with no stop at the frame. Framing: the edge of the door stays a defined line as it passes and what is beyond the threshold is legible within five frames of crossing. End: finishes inside the room with the doorway out of the frame behind and the subject in the middle distance.
Watch for. Untested on our stack. The frame filled by the door is a natural cut point, which is the safer use, because the model's geometry beyond the threshold often contradicts the room established before it. Count the doors, and generate the two sides separately if the interior has to match a plate.
From our notes
natural cut point
027Arc shot / 360 orbitorbit fully around the subjectprompt
Use it when
You need a subject to read as a solid object sitting in real space rather than a flat cut-out.
The phrasing
Arc shot / 360 orbit. Movement: the camera travels a full circle around the subject at a fixed radius, the subject holding its position on the floor while the background sweeps past behind it. Speed: one complete revolution across the length of the shot at a constant rate, no slowing at the start or the finish. Framing: the subject stays centred and the same size for the whole circle, so the background is the only thing that changes. End: back at the opening angle, subject the same size and on the same centre.
Watch for. Models often orbit part way and reverse, or rotate the subject instead of moving the camera. State the direction, the degree count and that the subject does not turn.
From our notes
reliable, our best DEPTH move
028Vertigo / dolly zoomdolly back while zooming inhybrid
Use it when
A character's ground shifts under them and the space has to distort without them moving.
The phrasing
Vertigo / dolly zoom. Movement: the camera retreats in a straight line along the lens axis away from the subject, the subject staying on its mark while the walls and floor slide past on both sides. Speed: about two metres of travel over four seconds at a constant rate, no pause at either end. Framing: the subject's head and shoulders stay on the same centre line, and the room stays visible on both sides so the later zoom has material to crop into. End: the subject at half its opening height with the whole room open around it.
Watch for. HYBRID: generate the dolly only. The counter-zoom is a post scale ramp keyed to the same frames, free and frame-exact; a model asked for both usually returns a plain push with a size change. Leave headroom on all four edges or the post zoom crops into nothing.
From our notes
models rarely nail it; do the zoom in post
029Red carpet trackingwalk backward ahead of a forward-moving subjectprompt
Use it when
A person has to arrive somewhere and the arrival is the event.
The phrasing
Red carpet tracking. Movement: the camera retreats along the subject's path, staying ahead of them at a constant distance with the lens turned back on their face, while the crowd and the barriers pass on both sides. Speed: matched to a walking pace, about 1.5 metres per second, held without closing or opening the gap. Framing: face and upper body stay centred, with the two lines of people readable down both edges. End: the same medium shot, subject one stride from the camera's stopping mark.
Watch for. The gap tends to collapse or open across the take. Give the distance as a number and say it does not change.
030Leading trackingwalk forward ahead, facing themprompt
Use it when
You are following someone through a route and their reactions carry the shot.
The phrasing
Leading tracking. Movement: the camera moves through the space in the same direction the subject walks, held two paces in front with the lens turned back on their face, and it turns with the path wherever the path turns. Speed: the subject's own walking pace, about 1.25 metres per second, each turn taking about one second. Framing: face and hands stay in frame, and the space they are walking away from stays readable behind their shoulders. End: framed on the face at the same distance, with the turn completed and the new corridor behind them.
Watch for. On a turn the model often swings to the subject's back. Say the lens stays on the face through the turn.
031Steadicam glidelow floating smooth interior movementprompt
Use it when
A space has to be shown as continuous, with one room leading into the next.
The phrasing
Steadicam glide. Movement: the camera travels through the interior at about 60 cm off the floor on a level line, passing doorways, table legs and furniture on both sides, with no vertical bob. Speed: unhurried walking pace, around one metre per second, held constant through the whole travel. Framing: the horizon stays level and the room's leading lines — door frames, counter edges, skirting — stay legible through every pass. End: settles in a doorway with the next room framed inside it.
Watch for. Low travel tends to drift upward to eye level. Give the height as a number and repeat it at the end of the beat.
032Low-angle hero walktrack low, looking up, to make them powerfulprompt
Use it when
A person has to read as larger than the space around them.
The phrasing
Low-angle hero walk. Movement: the camera retreats at knee height with the lens tilted up, keeping pace with the walking figure so their head stays against the ceiling or the sky behind them. Speed: walking pace, about 1.4 metres per second, constant, with no acceleration into the final frames. Framing: the head stays clear of the top edge, the jaw and eyes stay readable, and the vertical lines of the building converge behind the figure. End: the figure filling about three quarters of the frame height, still seen from below.
Watch for. Low angles crop the head. Say the top of the head stays inside the frame for the whole travel.
033High-angle vulnerabilitystraight down, isolatingprompt
Use it when
Someone has to look small inside the space they are standing in.
The phrasing
High-angle vulnerability. Movement: the camera hangs directly over the subject with the lens pointed straight down, and climbs vertically away from them while the floor plane stays flat to the lens. Speed: about three metres of climb over six seconds at a constant rate. Framing: the subject stays at the centre of the frame and stays large enough to read as a person — head, shoulders, and the shadow they cast on the floor. End: the subject at about a sixth of the frame height with empty floor on all four sides.
Watch for. The lens tends to tip off vertical as it climbs and a wall enters frame. Say the lens stays perpendicular to the floor for the whole climb.
034Crane / jib revealrise up and over to show a landscapeprompt
Use it when
The scale of a place is the point and the viewer should discover it, not open on it.
The phrasing
Crane / jib reveal. Movement: the camera rises from head height and travels forward across a foreground edge — a parapet, a ridge, a wall — so the near ground drops away and the landscape behind it enters from the bottom of frame. Speed: about eight metres of climb over six seconds, with the forward travel matched to the climb at a constant rate. Framing: the foreground edge stays in the lower third for the whole move, so the reveal reads as crossing something rather than cutting. End: a wide view with the foreground edge along the bottom and the landscape filling the rest of the frame.
Watch for. Models often climb without travelling forward, so nothing is revealed. Give the climb and the forward distance as two separate numbers.
035Silhouette revealpan from bright background to dark foregroundprompt
Use it when
A person should be identified by shape before the face is given away.
The phrasing
Silhouette reveal. Movement: the camera pans across the scene from a bright opening — window, doorway, sky — onto a figure standing between the lens and that light, so the figure enters from the frame edge and blocks the source. Speed: an even pan of about 30 degrees over three seconds, carrying a few degrees past the figure before it settles. Framing: the outline holds — shoulders, hairline, the gap between arm and body — hard against the bright field; nothing inside the outline needs detail. End: the figure filling one half of the frame as a flat dark shape, with the light source still open beside it.
Watch for. Models expose for the face and lift the shadow. Say the exposure sits on the bright background and the figure carries no interior detail.
036Rack focus slideslide sideways while shifting focusrisky
Use it when
Attention has to move between two things without a cut.
The phrasing
Rack focus slide. Movement: the camera slides sideways parallel to the subject plane while the plane of focus moves from the near object back to the one behind it. Speed: about one metre of slide over three seconds; the focus change starts at the halfway point and takes one second. Framing: both objects stay in frame for the whole slide, the near one on one third and the far one on the other, so sharpness is the only thing that changes. End: the far object sharp, the near one soft, both still in frame.
Watch for. RISKY: focus is the weakest control we have — models usually render both planes sharp or soften the whole frame. Test it on one short generation; if the plane does not move, generate it deep and pull the focus in post with a tracked blur.
From our notes
focus control is weak
037Dutch angle rolltilt off-centre for tensionprompt
Use it when
A scene has to tip without anything in it changing.
The phrasing
Dutch angle roll. Movement: the camera rotates around its own lens axis, carrying the horizon from level to about 12 degrees off, while the subject holds its position in the room. Speed: the roll runs at a constant rate across two seconds. Framing: the subject stays on the centre of the frame through the roll, and the corners stay clear of anything that has to be read, because the roll pushes content toward the edges. End: horizon 12 degrees off level, subject still centred.
Watch for. A roll is a free post transform on any clip with headroom, frame-exact and reversible. Generate it only when the light or the subject has to react to the tilt; otherwise the money buys nothing.
From our notes
or free in post
038Invisible cut shakethrow toward a colour block, pull back elsewherehybrid
Use it when
Two locations have to read as one continuous camera move.
The phrasing
Invisible cut shake. Movement: plate A — the camera swings hard toward a flat colour surface until that colour reaches every edge; plate B — the camera pulls back off an identically coloured surface in a second location. Speed: each swing covers about 90 degrees in a third of a second, accelerating, so the last frames are pure colour. Framing: the colour field must fill the frame for at least four frames on each plate, and the two colours must match in hue and brightness. End: plate A ends on full-frame colour; plate B opens off that colour onto the new location.
Watch for. HYBRID: generate the two plates only — the join is a post cut on the matching frames, and one generation asked for both sides will not give you a clean seam. Match the two colours by number; half a stop of difference makes the cut visible.
From our notes
two plates + POST cut
039Phone dropcamera drops onto a soft surfacerisky
Use it when
A shot has to feel like a phone someone actually put down.
The phrasing
Phone drop. Movement: the camera falls from about chest height onto a soft surface — bedding, a couch, grass — tumbling once on the way down so the room rotates past the lens, then coming to rest at an angle looking up. Speed: the fall lasts under a second and accelerates; the settle resolves within three frames of contact. Framing: one fixed element — a ceiling light, a window, a headboard — stays in frame through the tumble so the rotation is readable. End: a canted low view of the ceiling with the edge of the soft surface across the bottom of frame.
Watch for. RISKY: this is a look, not a camera path — models render a clean arc with no impact and the resting frame lands wherever they like. Real dropped-phone footage takes a minute to shoot; generating it usually costs more than it returns.
From our notes
a look, not a real move
040Perspective spinproduct spins one way, camera orbits the otherprompt
Use it when
A product has to show every side while the background does something the viewer cannot explain.
The phrasing
Perspective spin. Movement: the product turns on its own vertical axis in one direction while the camera orbits the product in the opposite direction, both around the same centre. Speed: the product turns 180 degrees and the camera travels 180 degrees the other way across four seconds, both at a constant rate. Framing: the product's face or label points at the lens for the first and the last second, and the product's size in frame does not change. End: label front again, with the background a half turn from where it started.
Watch for. Models often turn both in the same direction, which reads as a plain orbit. Name the two directions separately and give each its own degree count.
From our notes
excellent for product
041Match cut pushpush into an object, pull out of a different onehybrid
Use it when
Two unrelated things share a shape and the cut should argue they are the same thing.
The phrasing
Match cut push. Movement: plate A — the camera approaches an object along the lens axis until the object's outline fills the frame; plate B — the camera retreats from a second object that starts at the same size, in the same position. Speed: both travels run at the same rate, the object roughly doubling in frame height over two seconds. Framing: the shared shape — a circle, a rectangle, a silhouette — sits at the same size and the same place on plate A's last frame and plate B's first. End: plate B settles wide on the second object in its own space.
Watch for. HYBRID: generate the two pushes; the match is a post cut on the frames where the outlines align. Write the camera as approaching the object — asking for the object to come toward the lens makes it slide across the floor and the shape will not hold its position.
From our notes
two plates + POST cut
042Outfit toss swivelpan right on a throw, continue left in a new outfithybrid
Use it when
One person has to change wardrobe without a visible cut.
The phrasing
Outfit toss swivel. Movement: plate A — the camera whips right as the subject throws a garment across the lens, the garment filling the frame at the peak of the swing; plate B — the swing carries on in the same direction and decelerates onto the subject standing on the same mark in the second outfit. Speed: about 90 degrees of whip in eight frames, then half a second of deceleration into the new framing. Framing: the subject stands on the same mark at the same distance in both plates, and the frame passes through at least four unreadable blurred frames at the peak. End: the subject in the second outfit, same size and same position as the opening.
Watch for. HYBRID: generate the two whips; the join is a post cut inside the blurred frames. A single generation asked to change the outfit mid-take changes the face with it.
From our notes
the classic transition is POST
043Glitch shakelight shake mimicking a digital glitchfree in post
Use it when
A beat needs a break in the signal without anything in the scene changing.
What it should look like
Glitch shake. Movement: the whole frame jumps in short horizontal offsets of a few pixels with one or two vertical displacements, the content behind the jumps unchanged. Speed: six to ten jumps across half a second at irregular spacing, each lasting one or two frames. Framing: type and logos stay inside the safe area, because the offsets push content toward the edges. End: the frame back on its original centre.
Do this in your editor — it costs nothing

A keyframe in CapCut, Premiere, DaVinci or any editor does this. The command below is the same move for anyone who prefers a terminal — you do not need one.

ffmpeg -y -i in.mp4 -vf "scale=1134:2016:flags=lanczos,crop=1080:1920:x='(in_w-1080)/2+if(between(t,3.00,3.30),22*sin(1000*floor(t*20)),0)':y='(in_h-1920)/2+if(between(t,3.00,3.30),6*sin(1700*floor(t*20)),0)',rgbashift=rh=-8:bh=8:enable='between(t,3.00,3.30)',setsar=1" -c:v libx264 -crf 17 -preset medium -pix_fmt yuv420p -c:a copy out.mp4
What to change
3.00 and 3.30 are the glitch window and appear three times each — change all of them. 22 and 6 are the jump distance in pixels; a glitch is mostly horizontal, so keep the y value small. floor(t*20) is what makes it STEP rather than slide, and 20 (steps per second) is the single number that decides whether this reads as digital or as a wobble: 12 is chunky, 30 is frantic, above ~45 it turns to noise. The 1000 and 1700 multipliers inside sin() are pseudo-random seeds — change them to reshuffle which way each step jumps. rh=-8:bh=8 is the red/blue channel split in pixels; push to 16 for a heavy tear, 3 for a subtle one, and add gh= for a third channel. 1134:2016 is a 5% upscale for crop room and it caps the jump at 27px.
Why not generate it
Generating a glitch means describing a failure to a model built to avoid producing failures, and the result is either ignored or smeared across the whole clip instead of a 0.3s window. It also has to hit an exact window against the cut or the sound, which a generation cannot be aimed at. Everything here is a number you type, and you can move the window without re-rendering the shot.
Not run by us. This command is written from the documented filters, not executed here. Check it on a test clip before a client job.
Watch for. Do not generate this. It is done locally — per-frame offsets, a channel split and a one-frame repeat — free and frame-exact. A model asked for a glitch redraws the content between jumps, so the face and the type change while they shake.
044POV chest rig glidetrue first-personprompt
Use it when
The viewer has to be the person doing the thing, not watching them.
The phrasing
POV chest rig glide. Movement: the camera occupies the subject's eyeline and travels through the space as they move, their own forearms and hands entering from the lower edge when they reach for something. Speed: walking pace, about one metre per second, with a rise and fall of roughly two centimetres per step. Framing: the horizon sits near the middle of the frame, and whatever the person is about to touch stays visible before the hand reaches it. End: framed on the thing they have reached, hands in the lower third.
Watch for. The giveaway is a body in frame. Say that no head, shoulders or back of the subject is visible — only forearms and hands.
045Snack toss dropingredients dropped over an upward lensprompt
Use it when
Food or product has to arrive in frame with weight behind it.
The phrasing
Snack toss drop. Movement: the camera sits below the action looking up and climbs toward the source while the ingredients fall past the lens on both sides. Speed: the ingredients fall at full gravity in three staggered groups about six frames apart; the camera climbs about 20 cm over two seconds. Framing: the light source stays behind the falling pieces so their edges read, and the pieces pass either side of frame centre rather than covering it. End: the last group clearing the bottom of frame with the light open again at the top.
Watch for. Do not ask for the camera to be held while the food falls — stillness cannot be prompted; route a locked version to seedance 1.x cameraFixed. Models also drop everything in one clump; give the three groups and their frame spacing.
046Snorricamcamera rigged to the body; the world moves, not the personprompt
Use it when
A character is coming apart and the world should move instead of them.
The phrasing
Snorricam. Movement: the camera is fixed to the subject's torso at arm's length facing back at their face, so the subject keeps the same position and size in frame while the street, the corridor and everything behind them swings and slides past. Speed: the subject walks at normal pace, and the background passes at that same pace with a lurch on each step. Framing: the face stays centred and the same size for the whole take, and the background stays bright enough to read as moving. End: face still centred, background arrived somewhere different from where it started.
Watch for. Models treat this as ordinary tracking and let the subject drift in frame. Say the subject's position and size do not change and that all the motion belongs to the background.
047Clockwork spinlocked on a face while the frame rotatesprompt
Use it when
A face has to stay the fixed point while everything around it gives way.
The phrasing
Clockwork spin. Movement: the camera rotates around its lens axis with the face at the centre of rotation, so the face keeps its position while the room turns around it. Speed: 180 degrees over three seconds at a constant rate. Framing: the eyes stay on the centre of the frame through the rotation, and anything that has to be read stays inside a centred circle, because the corners lose content as the frame turns. End: the frame a half turn over, face still centred.
Watch for. A full-frame rotation is free in post on any clip with headroom. Generate it only when the light or the subject reacts to the turn; generated versions drift the face off the centre of rotation.
From our notes
or free in post
048Table slideslide across a counter past itemsprompt
Use it when
A spread of items has to be shown as one continuous inventory.
The phrasing
Table slide. Movement: the camera travels sideways just above a counter, parallel to its edge, passing plates, jars and hands at a constant height above the surface. Speed: about 60 cm over three seconds at a constant rate, with no ease at either end. Framing: the counter edge holds the same line across the lower third, and each item stays fully visible for at least half a second as it passes. End: framed on the last item with the passed items strung out behind it.
Watch for. Objects that cross close to the lens tend to render as a blur band. Keep the nearest item about 20 cm off the camera path, or name it as a foreground element the model should hold sharp.
049Top-down overhead slideflat down, moving parallelprompt
Use it when
A process or a layout is read best as a flat plan.
The phrasing
Top-down overhead slide. Movement: the camera looks straight down at the surface and travels parallel to it in one direction, staying the same distance above it for the whole move. Speed: about one metre over four seconds at a constant rate. Framing: the surface fills the frame with no wall or horizon entering, and the items stay square to the frame edges as they pass. End: framed on the arrangement at the far end of the travel, at the same height above the surface.
Watch for. Models tilt off vertical while travelling and a wall or a face enters the frame. Say the lens stays perpendicular to the surface for the whole move.
050Macro pushextremely close into textureprompt
Use it when
Material quality is the argument and the viewer has to see it up close.
The phrasing
Macro push. Movement: the camera approaches one detail of the object along the lens axis until that detail fills the frame, the object holding its position while the camera closes the distance. Speed: the last 10 cm taken over three seconds, decelerating into the final second. Framing: the whole object stays recognisable for the first second so the close-up is not read as a separate shot, and the named detail — grain, weave, crumb, condensation — stays centred. End: the detail filling the frame with the nearest surface sharp.
Watch for. At this distance the depth of field is a few millimetres. Name the one surface that must be sharp, or focus lands in the middle of the object and nothing reads.
From our notes
strong
051Steam rise pedestalup through rising steamprompt
Use it when
Hot food has to read as hot in the first two seconds.
The phrasing
Steam rise pedestal. Movement: the camera climbs vertically from just above the rim of the dish to a three-quarter height looking down, passing through the column of steam as it rises. Speed: about 40 cm of climb over four seconds at a constant rate, the steam lifting faster than the camera so the two rates separate. Framing: the steam stays lit from behind so it reads against the background, and the dish stays in the lower half of the frame through the climb. End: a three-quarter view of the whole dish with steam still crossing the upper half.
Watch for. Steam only reads against a dark or backlit field. Say where the light sits, or the model returns a grey haze that looks like noise.
052Unboxing pullstart inside the box, track back as the lid liftsprompt
Use it when
A product comes out of its own packaging and the box has to stay part of the shot.
The phrasing
Unboxing pull. Movement: the camera starts inside the open box level with the product and retreats straight back along the box's centre line as the lid swings up past the lens; box, product and table stay where they are. Speed: even retreat at about one box width per second, without ramping in or out. Framing: the product's front face stays centred and both inner box walls stay in frame for the whole pull. End: finishes on a three-quarter view of the open box with the product inside and a hand's width of table on each side.
Watch for. The move crosses from inside the box to the room, and our own measurement says one take holds one space — the interior can re-draw as a different box once the lid clears. Keep the retreat short enough that the box edges never fully leave frame, and compare the inner walls on the last frame against the first.
053Reveal from behind the platetilt up to show the settingprompt
Use it when
A dish is the subject and the room it sits in is the second half of the shot.
The phrasing
Reveal from behind the plate. Movement: the camera starts low behind the near rim of the plate with the rim filling the bottom edge, and tilts up on a fixed pivot until the table and the room behind it enter frame; the plate and everything on it stay where they sit. Speed: one continuous tilt at about 20 degrees per second, with no change of pace as the far rim crosses the centre. Framing: the plate rim stays in the lower third throughout and the food surface stays readable until the tilt passes it. End: finishes on the table with the plate in the bottom third and the setting filling the upper two thirds.
Watch for. Frame one contains only the plate, so the room arrives invented and can contradict the brief. Describe the room in the scene text as well as the dish, and check the first frame in which the walls appear.
054Diagonal slicecorner to corner across a flat-layprompt
Use it when
A flat-lay of several items needs motion without a cut.
The phrasing
Diagonal slice. Movement: the camera stays overhead with the lens pointed straight down and travels in a straight line from one corner of the flat-lay to the opposite corner; every object stays fixed on the surface. Speed: constant glide, the full diagonal crossed in 4 seconds, with no change of pace at the midpoint. Framing: at least three objects fully inside the frame edges at any moment, with the surface texture held sharp. End: finishes with the last corner object centred and the rest of the arrangement running off the opposite edge.
Watch for. Objects duplicate or change as they pass out of frame and back in. Count the items in the prompt and keep the diagonal short enough that nothing leaves and returns.
055Spiral up orbitorbit while liftingprompt
Use it when
A single product has to be read from several sides in one take.
The phrasing
Spiral up orbit. Movement: the camera circles the product on a fixed radius while climbing, rising about one product height per quarter turn; the product keeps its place on the surface and its facing throughout. Speed: continuous, three quarters of a turn over 5 seconds, the climb rate matched to the turn rate. Framing: the product stays centred and the surface edge stays visible on at least one side so the height reads. End: finishes at a high three-quarter angle, roughly 45 degrees above the surface, with the whole product and its shadow in frame.
Watch for. The shadow stays pinned to the opening camera angle while the camera climbs, which reads as the product floating. Check the shadow direction on the first and last frames.
056Glow-up panpan across an unlit product as lights come uprisky
Use it when
A product goes from dark to lit as the shot introduces it.
The phrasing
Glow-up pan. Movement: the camera pans sideways across the product from a fixed position while the key light rises from dark to full; the product does not move. Speed: one unbroken pan at about 12 degrees per second, the light reaching full at the pan's midpoint. Framing: the product silhouette readable against the background from the first frame, the label legible from the midpoint on. End: finishes with the lit product in the centre third and the background about two stops darker than the product.
Watch for. Lighting ramps drift — the light restarts, reverses, flickers or shifts colour temperature partway through, and the label can change as the exposure lifts. Expect to re-run, and if two takes both drift, generate the pan under a steady key and build the ramp locally.
From our notes
lighting changes drift
057Ground skimtrack an inch off the asphalt behind a runner or boardprompt
Use it when
A runner, skater or rider needs speed felt from ground level.
The phrasing
Ground skim. Movement: the camera holds a constant height of a few centimetres above the asphalt and tracks forward behind the subject at the subject's own speed, so the gap between lens and subject stays the same while the road surface runs past underneath. Speed: fast — the asphalt streaking past, the subject holding one place in frame. Framing: the subject's feet or the board's underside stay in frame along with about two metres of road ahead of them. End: finishes at the same gap, subject in the lower centre, the road running out to the horizon above it.
Watch for. Written loosely this comes back near-still: the camera floats and the ground barely moves. Name the ground texture passing and the stride or wheel rate rather than writing the word fast on its own.
058Whip-around reversaltrack from the front, sweep around to behindrisky
Use it when
The shot has to change sides on a moving subject without a cut.
The phrasing
Whip-around reversal. Movement: the camera leads the subject from the front, then arcs around one side on a constant radius until it is directly behind them on the same line of travel; the subject keeps walking straight ahead through the whole arc. Speed: the arc takes about 1.5 seconds and runs faster through the side than at either end. Framing: the subject's head and shoulders stay in frame for the whole arc and the background stays readable as it swings past. End: finishes behind the subject at shoulder height with their back centred.
Watch for. Identity drifts as the camera crosses the profile — the face can come back as a different person and hair and wardrobe reset on the far side. Lock the subject with an identity plate, and with two people in shot expect the pair to swap features mid-arc.
059Low-to-high jump followstart low, lift as they jumpprompt
Use it when
A jump, dunk or vault is the beat of the shot.
The phrasing
Low-to-high jump follow. Movement: the camera starts at ankle height a stride in front of the subject and rises vertically as the subject leaves the ground, reaching the subject's chest height at the top of the jump; the camera's position on the ground does not change. Speed: the lift is tied to the jump — slow through the crouch, quick through the push off, easing as the subject reaches the apex. Framing: the subject's feet in frame at the start and their whole body in frame at the apex. End: finishes at chest height with the subject centred and the ground about a body length below the bottom edge.
Watch for. The jump and the lift come back out of phase, with the camera already high before the feet leave the ground. Write the crouch, the push and the apex as three timed beats rather than one word.
060FPV drone diveswoop down a structureprompt
Use it when
A building, cliff or tower is being introduced with scale.
The phrasing
FPV drone dive. Movement: the camera starts above the top of the structure and descends close to its face, following the wall down, then pulls out of the descent near the base into level forward flight; the structure stays fixed. Speed: accelerating through the drop and decelerating into the level-out, the whole dive over about 4 seconds. Framing: the face of the structure fills one side of frame through the descent, with its windows or panel lines readable rather than smeared. End: finishes in level forward flight about one storey above the ground, the base of the structure passing on one side.
Watch for. Lettering on the building garbles at speed — small text is a measured failure on every provider we use. Keep signage out of the framing, or generate it as its own slower shot.
061Horizon lock rollcamera rotates while the horizon stays levelfree in post
Use it when
A subject turns against a level world for a stylised beat.
What it should look like
Horizon lock roll. Movement: the frame rotates around the lens axis while the horizon stays flat across the frame, so the subject and its background turn together against a level world. Speed: one full rotation over 6 seconds at a constant rate, with no change of pace at the quarter turns. Framing: the horizon stays a straight line across the middle of frame and the subject stays centred on the axis of rotation. End: finishes upright with the subject centred and the horizon level.
Do this in your editor — it costs nothing

A keyframe in CapCut, Premiere, DaVinci or any editor does this. The command below is the same move for anyone who prefers a terminal — you do not need one.

ffmpeg -y -i in.mp4 -vf "scale=1350:2400:flags=lanczos,rotate='0.14*sin(2*PI*t/8)':ow=1350:oh=2400:c=black,crop=1080:1920:(in_w-1080)/2:(in_h-1920)/2,setsar=1" -c:v libx264 -crf 17 -preset medium -pix_fmt yuv420p -c:a copy out.mp4
What to change
0.14 is the peak roll in RADIANS (8 degrees) — the rotate filter does not take degrees, so for degrees write '8*sin(2*PI*t/8)*PI/180'. The 8 in t/8 is the period in seconds: one full sway out and back every 8s. Replace the sin with a one-way settle — rotate='0.14*min(t/3,1)' — for a single roll instead of a sway. THE COVERAGE MATH IS THE TRAP: for a centred w x h crop out of a W x H source at angle A, you need W >= w*cos(A)+h*sin(A) and H >= h*cos(A)+w*sin(A). At 1080x1920 out of 1350x2400 and 8 degrees that is 1336 <= 1350 and 2051 <= 2400 — it fits with 14px to spare. Raise the angle and you MUST raise 1350:2400 with it or black wedges appear in the corners; c=black is chosen so they are obvious rather than hidden. To counter-roll a source that is already rolling, negate the expression and match its period. The 1.25x upscale softens — a 2160x3840 source scaled down to 1350x2400 stays sharp.
Why not generate it
A roll is one of the few moves a model will sometimes render, and it will not hold the horizon while it does — that is the entire point of the move and it is exactly the part that gets lost. Doing it here means the angle is a number and the period is a number, so 'half as much' costs a re-render rather than another generation and another audit.
Not run by us. This command is written from the documented filters, not executed here. Check it on a test clip before a client job.
Watch for. Do not generate this — it is a rotation applied to a finished frame and it costs nothing locally. Generate a clean tracking plate with headroom on all four sides, because the rotation crops the corners, then roll and re-level it in post.
062Bicycle leadrear-mounted, filming who chases behindprompt
Use it when
A chase or a group ride is shot from inside the pack.
The phrasing
Bicycle lead. Movement: the camera is mounted on the back of the lead bicycle facing backward, holding a fixed distance and angle to the riders following, so the road, kerb and buildings recede while the chasing riders hold about the same size in frame. Speed: the world recedes at riding pace and small bumps from the road surface come through the mount. Framing: the nearest chasing rider's face and front wheel stay in frame, with the road between the two bikes visible. End: finishes with the chasing rider slightly larger than at the start, centred, the road running away behind them.
Watch for. The rig gets generated as a visible object rather than as the camera position, putting a bicycle in shot. State that the camera is the mount, not a bicycle seen in frame.
063Parallel speed matchpan at the exact speed of a passing carprompt
Use it when
A vehicle passes and its surface detail has to stay sharp.
The phrasing
Parallel speed match. Movement: the camera stays in one position and pans to follow the passing car, the pan rate matched to the car's angular speed so the car keeps the same place in frame while the background sweeps behind it. Speed: the pan neither lags nor overtakes the car, and background detail streaks horizontally throughout. Framing: the car's full length in frame with its wheels and body line sharp against the smeared background. End: finishes on a three-quarter rear view as the car continues out toward the far edge of frame.
Watch for. The match slips and the car slides forward or backward in frame, and the wheels often render as static discs rather than turning. Write the wheel rotation as its own movement, separate from the pan.
064Chaotically handheldraw documentary shakeprompt
Use it when
The piece needs a documentary or first-person register.
The phrasing
Chaotically handheld. Movement: the camera is carried by hand and hunts around a fixed world — small left-right corrections, a vertical drift on each step, the subject sliding off centre and being pulled back. Speed: several small corrections per second with a larger re-frame about every two seconds, and no smooth arcs anywhere. Framing: the subject's face stays inside the middle half of frame while the frame hunts, with the horizon tilting up to about 8 degrees either way. End: finishes with the subject back near centre and the frame still hunting.
Watch for. Handheld written as an adjective produces a smooth glide with a light wobble. Spell out the corrections, the step drift and the tilt range as separate movements, and never pair it with a request to settle.
065Infinity revealpan over a ridge to a hidden valleyprompt
Use it when
A landscape is withheld and then given all at once.
The phrasing
Infinity reveal. Movement: the camera rises over the ridge line from behind it while travelling forward, so the ground falls away and the valley enters from the bottom of frame; ridge and valley stay fixed. Speed: a steady forward creep, the valley taking about 3 seconds to fill the lower half of frame. Framing: the ridge line readable as a hard edge until it exits the bottom of frame, and the far side of the valley holding the upper third. End: finishes on a wide of the valley with the near ridge out of frame and the far ridge on the horizon.
Watch for. The valley is not in frame one, so it is invented — its scale, season and weather can contradict the ridge it appears from. Describe the valley in the scene text, not only the ridge.
066Window view passpush out through a car or train windowprompt
Use it when
A journey shot moves from the passenger's seat to the view outside.
The phrasing
Window view pass. Movement: the camera starts inside the vehicle with the window frame around the edges of shot and travels forward through the open window into the open air, the window frame passing the lens and leaving at the edges as the landscape opens out. Speed: an even push, the window frame clearing the lens in about 2 seconds, then holding forward travel alongside the vehicle. Framing: the inner edge of the window readable at the start and the passing landscape readable once through. End: finishes outside and level with the moving vehicle, landscape filling the frame and the vehicle's body along one edge.
Watch for. This crosses two spaces in one take, which our own measurement says is where a take breaks — the interior or the vehicle can re-draw as a different one once the lens is outside. Count the spaces before ordering it, and consider two shots joined on the window frame.
067Drone horizon tiltfly forward, tilt up to the sunprompt
Use it when
A travel or landscape sequence needs an opening or closing beat.
The phrasing
Drone horizon tilt. Movement: the camera flies forward at a constant height while tilting up from the ground ahead to the sky, so the land slides down out of frame and the sun enters from the top; the terrain stays fixed below. Speed: forward flight even throughout, the tilt taking about 3 seconds and running in the second half of the move. Framing: the horizon stays in frame for the whole tilt, travelling from the upper third to the lower third, and the sun reads as a disc rather than a blown field. End: finishes with the horizon in the bottom quarter and the sun in the upper centre, land still passing below.
Watch for. Pointing into the sun blooms the whole frame and flattens the land to a silhouette. Ask for the sun low and partly occluded, and check the last second for clipped highlights.
068Under-bridge shadow passtrack through shadow into sunlightprompt
Use it when
A drive or a walk needs a mood change without a cut.
The phrasing
Under-bridge shadow pass. Movement: the camera tracks forward in a straight line from the shadow under the bridge out into open sunlight, the bridge deck passing overhead and the exposure lifting as the camera clears it. Speed: constant forward travel, the shadow edge crossing the lens about 2 seconds in. Framing: the underside of the bridge readable while in shadow, the road ahead readable once out, and the shadow edge visible as a hard line across the ground. End: finishes in full sunlight with the bridge behind and out of frame and the road running to the centre of frame.
Watch for. The exposure change is the point of the move and it is also what drifts — the shift arrives early, reverses, or carries a colour cast that does not match the daylight beyond. If two takes both drift, generate under one exposure and grade the lift locally.
From our notes
strong MOOD change
069Foreground tree sweeptrack behind trees passing rhythmicallyprompt
Use it when
A landscape or a walking subject needs depth without a 3D build.
The phrasing
Foreground tree sweep. Movement: the camera tracks sideways past a line of trees standing between it and the subject, each trunk crossing the lens in turn and wiping the frame; the trees and the subject both stay where they are and only the camera travels. Speed: even sideways travel with a trunk crossing frame about every 0.8 seconds. Framing: the subject beyond the trees holds the same place in frame and stays readable between trunks, with the near trunks out of focus as they cross. End: finishes past the last trunk on a clear view of the subject, centred.
Watch for. The parallax is the whole value and it is the part that collapses — trunks and subject can travel at the same rate, which reads as a flat cut-out. Compare two frames a second apart and measure how far the trunk moved against how far the subject moved.
From our notes
best cheap DEPTH move
070Water reflection tiltstart on a puddle, tilt up to the real viewprompt
Use it when
Rain or wet ground is on location and the reveal can use it.
The phrasing
Water reflection tilt. Movement: the camera starts pointed down at the puddle holding the reflection and tilts up on a fixed pivot, past the far edge of the water, to the subject itself; the puddle and the subject stay in place. Speed: one continuous tilt at about 15 degrees per second, with no change of pace as the waterline crosses the frame. Framing: the reflected subject readable in the water at the start, the waterline crossing frame through the middle of the move, the subject readable at the end. End: finishes with the subject centred and a strip of the puddle in the bottom eighth of frame.
Watch for. The reflection and the subject do not match — lettering reads correctly in both instead of reversing, and the pose differs. Ask for a soft, broken water surface and check the two against each other before accepting the take.
071180° vista panslow panoramic arcprompt
Use it when
A location needs to be established in its full width.
The phrasing
180° vista pan. Movement: the camera stays in one position and pans through half a turn across the view, the landscape crossing the frame from one edge to the other; nothing in the scene changes position. Speed: the half turn takes about 10 seconds at a constant rate, roughly 18 degrees per second. Framing: the horizon holds the same height across the whole pan, with at least one near landmark in frame at any moment to carry the scale. End: finishes 180 degrees from the start, with the view that began behind the camera now filling the frame.
Watch for. A long pan re-draws the scene as it enters frame, so the far half of the arc can belong to a different place than the near half. Keep the arc under about 10 seconds of new material, or break it into two shots.
072Vortex rollrotate 360° while walking forward down a hallwayprompt
Use it when
A corridor or tunnel sequence needs a disorienting beat.
The phrasing
Vortex roll. Movement: the camera travels forward down the centre of the hallway while rolling once around the lens axis, so the walls, floor and ceiling turn past one another; the hallway itself stays fixed. Speed: the forward walk is even and the full roll takes about 4 seconds at a constant rate across it. Framing: the far end of the hallway stays at the centre of the roll and the wall lines stay readable as they turn. End: finishes upright at the end of the hallway with the doorway centred.
Watch for. The roll and the walk fight each other and the model delivers one or the other. Write them as two simultaneous movements each with its own rate, and check that the vanishing point did not wander off centre.
073Vertigo tilttilt up while moving downhybrid
Use it when
A realisation or a shift in a character's footing needs a visual.
The phrasing
Vertigo tilt. Movement: the camera descends vertically while tilting up toward the subject, so the subject holds its size and place in frame while the space behind it stretches; the subject and the background stay where they are. Speed: the descent runs evenly over about 4 seconds with the tilt rate tied to it, so the subject's height in frame does not change. Framing: the subject's full height in frame for the whole move and the background behind it readable as it stretches. End: finishes low, looking up at the subject, the background standing taller behind it than at the start.
Watch for. Generate the descent and the up-tilt only; the counter-zoom half — the focal length change that holds the subject's size while the background stretches — is not a parameter any provider exposes, so build it locally as a scaling ramp on the plate. Generate with headroom on all four sides so the ramp has material to take.
074Dreamy driftultra-slow handheld float, shallow focusprompt
Use it when
A memory, a dream, or a slow product register.
The phrasing
Dreamy drift. Movement: the camera floats forward and sideways at once along a shallow curve, with a small vertical drift on the operator's breathing; the scene stays fixed and the frame slides across it. Speed: slow — the frame crosses about a tenth of its own width per second, with no sharp corrections. Framing: the subject sits in the plane of focus with foreground and background soft, and stays inside the middle third as the frame drifts. End: finishes with the subject just off centre and the drift still running.
Watch for. Slow travel plus shallow focus is the pairing that most often comes back as a held frame with grain, and grain reads as motion on a coarse pixel-difference measurement. Run a structural dead-air check on the result, and never write the word still anywhere in the prompt.
075Mirror maze swivelpan between mirrors until orientation is lostrisky
Use it when
A set with mirrors or glass and a deliberately lost orientation.
The phrasing
Mirror maze swivel. Movement: the camera pans on a fixed pivot from one mirror panel to the next, each panel carrying a reflection of the panels behind it, so the pan crosses real space and reflected space in turn; the panels stay fixed. Speed: about 25 degrees per second, slowing to roughly 8 degrees per second as each seam crosses the centre and picking up again after it, four panels in about 6 seconds. Framing: the subject's reflection readable in at least one panel at every moment, with the seams between panels visible as vertical lines. End: finishes on a panel carrying the subject twice, once direct and once reflected.
Watch for. Reflections drift — a reflected figure gains clothing the real one does not have, faces differ between the two, and a reflection appears with no source in frame. Count the panels and the figures explicitly, and check every reflection in motion, because this passes on a still and fails across the sequence.
076Upside-down trackinverted, tracking the ceiling lineprompt
Use it when
A corridor, ceiling or interior needs an unsettled register.
The phrasing
Upside-down track. Movement: the camera is inverted and travels forward along the corridor with the ceiling line running through the frame and the floor above it; the room stays fixed and only the camera travels. Speed: even forward travel with a ceiling fixture passing the lens about every 2 seconds. Framing: the ceiling line stays a straight edge across frame and the far end of the corridor stays at the vanishing point. End: finishes with the far doorway centred and the frame still inverted.
Watch for. The inversion is a free 180-degree rotation on an ordinary forward-tracking plate, so generate it upright unless the scene carries something the flip would break — readable signage, falling liquid, hair or smoke. If any of those are in shot, generate inverted and check that gravity still runs the right way.
From our notes
or free in post
077Ghostly dragslow shutter motion trailsfree in post
Use it when
A moving figure or object needs to read as speed and after-image rather than as a clean pass.
What it should look like
Ghostly drag. Movement: every moving subject leaves a smeared trail along the path it has just travelled, the trail running one to three frames behind the solid edge, while the frame keeps the same world position. Speed: trail length scales with subject speed — a walking figure drags about a body width, a hand gesture about a palm width, an object at rest drags nothing. Framing: the subject's head and shoulders stay solid enough to read as one person, with the trail behind the silhouette and never across the face. End: settles on a frame where the trail has collapsed back into the subject and the silhouette has a single edge.
Do this in your editor — it costs nothing

A keyframe in CapCut, Premiere, DaVinci or any editor does this. The command below is the same move for anyone who prefers a terminal — you do not need one.

ffmpeg -y -i in.mp4 -vf "tmix=frames=9:weights='1 1 2 3 5 8 13 21 34',setsar=1" -c:v libx264 -crf 17 -preset medium -pix_fmt yuv420p -c:a copy out.mp4
What to change
frames=9 is the trail LENGTH in frames — at 30fps that is a 0.3s smear. 5 is a subtle shutter drag, 15 is a full ghost, above about 24 it turns to mush and the render slows noticeably. The weights list must contain exactly as many numbers as frames=, and it is the trail SHAPE: this one is a Fibonacci ramp so one end dominates and the other fades off. Flat weights ('1 1 1 1 1 1 1 1 1') give an even blur with no direction and no sense of time. CHECK THE ORDER ON YOUR OWN BUILD FIRST: render two seconds and look at whether the smear trails BEHIND the moving subject. If the ghost leads the subject, reverse the list to '34 21 13 8 5 3 2 1 1'. For bright objects on a dark ground, lagfun=decay=0.94 is the better filter — it keeps light trails without dimming the whole frame the way frame-averaging does.
Why not generate it
This is a shutter effect, not a camera move, and no provider exposes shutter angle. Prompting for motion trails gets you a model's idea of a blurry frame, which is uniform softness on everything including the parts that were not moving — the opposite of what a long exposure does. Free here, and non-destructive: you can dial the trail length ten times in the time one generation takes.
Not run by us. This command is written from the documented filters, not executed here. Check it on a test clip before a client job.
Watch for. Do not generate this. It is a local frame-blend — stack 3 to 5 neighbouring frames with falling opacity in the post timeline; a model asked for slow shutter returns generic motion blur it cannot control. The fixed frame under it comes from the fixed-camera parameter or from the timeline, never from words in a prompt.
078In-camera warp zoomjerk back at the end as a transition stretchfree in post
Use it when
Two shots need to be joined with a jolt instead of a straight cut.
What it should look like
In-camera warp zoom. Movement: the frame stretches outward from its own centre over the last frames of the outgoing shot, then snaps back one step past its starting scale on the frame of the cut. Speed: the stretch builds across five to seven frames and the snap back happens inside one; the whole figure occupies under half a second. Framing: the centre third stays readable through the stretch so the viewer keeps hold of the subject while the outer edges smear. End: lands on the first frame of the incoming shot at its own scale, with no stretch left by frame three.
Do this in your editor — it costs nothing

A keyframe in CapCut, Premiere, DaVinci or any editor does this. The command below is the same move for anyone who prefers a terminal — you do not need one.

ffmpeg -y -i in.mp4 -vf "zoompan=z='1.15+0.30*(clip(on/96,0,1)*clip(on/96,0,1)*(3-2*clip(on/96,0,1)))-0.43*(1-pow(1-clip((on-96)/24,0,1),3))':x='iw/2-(iw/zoom/2)':y='ih/2-(ih/zoom/2)':d=1:s=1080x1920:fps=30,setsar=1" -c:v libx264 -crf 17 -preset medium -pix_fmt yuv420p -c:a copy out.mp4
What to change
1.15 = the baseline zoom the whole shot sits at. It has to be above 1.0 because zoompan cannot go below the source frame, and that headroom is what the snap-back falls into — raise it to 1.25 for a bigger jerk and give the source more resolution to pay for it. 0.30 = how far it pushes in before the snap (ends at 1.45). 0.43 = how far the snap gives back (lands at 1.02). 96 = the frame the snap fires on; 24 = how many frames it takes (3.2s in, 0.8s of snap at 30fps). Change 96 in all four places. pow(1-J,3) is the ease-out that makes it a jerk rather than a glide — exponent 3 is a hard snap, 2 is softer, 5 is violent. Cut on the last frame; the snap IS the transition.
Why not generate it
Every attempt to prompt this comes back as a smooth zoom, because a model renders continuous motion and the whole point here is the discontinuity. It is also a transition, which means it has to land on an exact frame against the cut — a generation cannot be aimed at a frame, and re-rolling until one lands close is how a $1.20 call becomes six.
Not run by us. This command is written from the documented filters, not executed here. Check it on a test clip before a client job.
Watch for. Do not generate this. It is a local scale-and-blur ramp across the cut point, built on two finished shots; a model given "warp zoom" returns a plain push and wastes the call. Keep the stretch under half a second — past that it reads as a render fault rather than a transition.
079Ken Burns driftslight digital scale-up on a static shotfree in post
Use it when
A still photograph or a held plate has to sit inside a cut without going dead.
What it should look like
Ken Burns drift. Movement: the crop rectangle travels across a still image, scaling up between 4 and 8 percent over the shot while sliding toward the point of interest. Speed: constant, about 0.3 pixels of subject growth per frame at 1080p — enough that no frame repeats itself, slow enough that nobody names the move. Framing: the subject stays inside the crop at every scale, with its edges at least 40 pixels clear of frame. End: settles on a tighter crop centred on the point of interest, holding the same headroom it opened with.
Do this in your editor — it costs nothing

A keyframe in CapCut, Premiere, DaVinci or any editor does this. The command below is the same move for anyone who prefers a terminal — you do not need one.

ffmpeg -y -i in.mp4 -vf "zoompan=z='1+0.1364*(clip(on/120,0,1)*clip(on/120,0,1)*(3-2*clip(on/120,0,1)))':x='(iw-iw/zoom)*(0.5-0.25*(clip(on/120,0,1)*clip(on/120,0,1)*(3-2*clip(on/120,0,1))))':y='(ih-ih/zoom)*(0.5-0.18*(clip(on/120,0,1)*clip(on/120,0,1)*(3-2*clip(on/120,0,1))))':d=1:s=1080x1920:fps=30,setsar=1" -c:v libx264 -crf 17 -preset medium -pix_fmt yuv420p -c:a copy out.mp4
What to change
120 = the move's length in FRAMES (4s x 30fps). It appears six times; change all six together. If the clip is longer the move finishes and holds, because clip() caps progress at 1 — that is safe, not a bug. 0.1364 = the zoom travel, computed as 1/(1-A)-1 where A is the fraction of frame consumed; A=0.12 gives 0.1364, A=0.20 gives 0.2500, A=0.30 gives 0.4286. Above A≈0.35 on a 1080p source it visibly softens. 0.25 and 0.18 are how far the frame slides left and up as it grows, as a fraction of the available slack — set both to 0 for a pure centred push. s=1080x1920 and fps=30 are the delivery size and rate; feed this a 2160x3840 source and deliver 1080x1920, because a moving crop on a source the same size as the output is an upscale and it softens. The clip(...)*clip(...)*(3-2*clip(...)) block is smoothstep easing — replace the whole block with clip(on/120,0,1) for a linear move, which will look mechanical.
Why not generate it
Generating this means paying a video model to render four seconds it has no reason to change, and it will not hold the frame while it does — stillness cannot be prompted, so you get drift you did not ask for on top of the drift you did. Here the move is exact, the timing is exact, and a revision costs another thirty seconds of CPU instead of another render.
Watch for. Do not generate this. It is a crop-and-scale keyframe pair on a still, done locally for nothing. The measured trap is under-doing it: a push below about 0.3 pixels per frame is sub-pixel and the frame is structurally identical to itself a second later, which our dead-air check reads as a held frame.
080Speed ramp pushslow in, fast middle, slow outfree in post
Use it when
A push needs to carry emphasis at one point in the line rather than a flat approach.
What it should look like
Speed ramp push. Movement: a push along the camera's own axis toward the subject, where the rate of travel changes through the shot and the path does not. Speed: the first fifth of the distance takes half the running time, the middle three fifths take a quarter, the last fifth takes the rest. Framing: the subject's eyeline stays on the same horizontal third throughout, so the acceleration reads as distance closing rather than as a reframe. End: finishes in a chest-up frame from a waist-up frame, on the same axis and the same third it started on.
Do this in your editor — it costs nothing

A keyframe in CapCut, Premiere, DaVinci or any editor does this. The command below is the same move for anyone who prefers a terminal — you do not need one.

ffmpeg -y -i in.mp4 -filter_complex "[0:v]trim=0:1.5,setpts=(PTS-STARTPTS)/0.5[a];[0:v]trim=1.5:3.0,setpts=(PTS-STARTPTS)/2.5[b];[0:v]trim=3.0:5.0,setpts=(PTS-STARTPTS)/0.5[c];[a][b][c]concat=n=3:v=1:a=0,fps=30,setsar=1[v]" -map "[v]" -an -c:v libx264 -crf 17 -preset medium -pix_fmt yuv420p out.mp4
What to change
The three trim ranges are the segment boundaries in SOURCE seconds — 0:1.5, 1.5:3.0, 3.0:5.0. They must be contiguous, and the last one must not exceed the clip's real duration or the tail is silently dropped. The divisors are the speed: /0.5 is half speed, /2.5 is 2.5x, /1 is untouched. Output length per segment = source length / divisor, so this 5s clip becomes 3.0 + 0.6 + 4.0 = 7.6s. Add more [d];[e] segments and raise concat=n= to match — five segments at 0.4 / 0.8 / 2.5 / 0.8 / 0.4 reads far less mechanical than three. Audio is dropped with -an on purpose: ramping a dialogue track needs pitch handling, not this. Slow segments below about /0.5 start to judder because there are no frames to fill the gap; you can add minterpolate=fps=60:mi_mode=mci to the slow segments only, but our own measurement on generated clips is that synthesised frames come back 25-33% softer on the moving pixels than the real frames either side, which installs a visible sharpness flicker on hair, teeth and fast edges.
Why not generate it
A model cannot be told to change speed partway through a shot with any accuracy, and the ramp usually has to land against a music beat, which means frame accuracy a generation cannot give you. Generate the move once at a constant rate and the ramp is free; generate the ramp and you are paying per attempt to hit a timing you could have typed.
Not run by us. This command is written from the documented filters, not executed here. Check it on a test clip before a client job.
Watch for. Do not generate this ramp. Generate a push at one constant rate, then retime it locally — a model cannot be trusted to place an acceleration, and paying for a second take to fix the curve doubles the cost of a free edit. Retiming below 40 percent speed needs optical-flow interpolation or the slow ends judder.
081Hyperlapse orbitstepped stills stitched into a fast orbitfree in post
Use it when
A static object or location needs a fast circuit built from stills rather than a generated move.
What it should look like
Hyperlapse orbit. Movement: the camera travels a circular path around the subject in discrete steps, each step its own still, with the subject anchored at the same point in the world for every one. Speed: 6 to 12 degrees of arc between frames, so a half circle runs 15 to 30 frames; the background sweeps past in blocks while the subject holds. Framing: the subject's centre mass stays within 20 pixels of the same frame coordinate across every step, or the orbit reads as a shake. End: settles at the chosen arc angle — most often a quarter or a half turn from the opening — with the subject still on centre.
Do this in your editor — it costs nothing

A keyframe in CapCut, Premiere, DaVinci or any editor does this. The command below is the same move for anyone who prefers a terminal — you do not need one.

ffmpeg -y -i in.mp4 -vf "select='not(mod(n,6))',setpts=N/FRAME_RATE/TB,tmix=frames=3,fps=30,setsar=1" -an -c:v libx264 -crf 17 -preset medium -pix_fmt yuv420p out.mp4
What to change
6 is the speed multiplier and the whole effect: keep every 6th frame, so a 12s orbit becomes 2s. 4 is brisk, 10 is aggressive, above about 15 the subject jumps too far between kept frames to read as continuous. tmix=frames=3 is the smear that sells it as a hyperlapse rather than a dropped-frame stutter — set it to 1 or remove it for the hard stepped look, 5 for a heavier streak. Order matters: tmix runs AFTER the select, so it averages the KEPT frames, which is what produces the long-exposure feel rather than a blur of frames you threw away. fps=30 fixes the output rate. -an is required because the audio is not retimed and would run 6x long. HONEST LIMIT: the orbit must already exist in the source — ffmpeg cannot walk a 2D frame around a subject. This recipe supplies the speed and the stepped texture, nothing else.
Why not generate it
The orbit path is the expensive half, and it is the half you should not buy either — render it in Blender, where it is free and frame-exact, or shoot a real walk-around. The stepped hyperlapse LOOK is pure post and models will not produce it, because they are built to render smooth motion: ask for 'stepped' or 'time-lapse' and you get a smooth orbit with the colour of a time-lapse. You would be paying a generation fee for frame selection.
Not run by us. This command is written from the documented filters, not executed here. Check it on a test clip before a client job.
Watch for. Do not generate this. It is stitched locally from a set of stills or from a decimated clip, then stabilised on the subject. The stabilisation is the whole job: without an anchor point tracked across every step the result is a judder, not an orbit.
082Digital jitterartificial micro-shakefree in post
Use it when
A clean digital frame needs handheld texture or a glitch beat under it.
What it should look like
Digital jitter. Movement: the whole frame offsets by 1 to 4 pixels in a new direction each frame, with no drift accumulating across the shot. Speed: a fresh offset every frame at 24 fps; halve the amplitude and hold each offset for two frames to get a coarser jitter. Framing: any on-screen type sits at least 60 pixels inside the safe area so no offset pushes a letter past the edge. End: the amplitude falls to zero across the last eight frames and the frame sits back on its original coordinate.
Do this in your editor — it costs nothing

A keyframe in CapCut, Premiere, DaVinci or any editor does this. The command below is the same move for anyone who prefers a terminal — you do not need one.

ffmpeg -y -i in.mp4 -vf "scale=1134:2016:flags=lanczos,crop=1080:1920:x='(in_w-1080)/2+3.0*sin(2*PI*11.3*t)+2.0*sin(2*PI*7.1*t+2.0)':y='(in_h-1920)/2+2.5*sin(2*PI*9.7*t+0.7)+1.8*sin(2*PI*5.3*t+1.4)',setsar=1" -c:v libx264 -crf 17 -preset medium -pix_fmt yuv420p -c:a copy out.mp4
What to change
3.0 / 2.0 / 2.5 / 1.8 are the four amplitudes in pixels. Keep the total under about 6px or it stops reading as a camera and starts reading as a fault. 11.3, 7.1, 9.7, 5.3 are the frequencies in Hz — they are deliberately not whole numbers and not multiples of each other, because two sines at 8 and 16 Hz produce a visibly repeating loop. The +2.0, +0.7, +1.4 phase offsets stop the two axes starting together. 1134:2016 is a 5% upscale for crop room; with amplitudes this small you could instead crop 1080x1920 out of a 1152x2048 source and skip the scale entirely. Below about 0.3px per frame nobody sees it and the frame is effectively dead, so a dead-air check will still fail the shot — keep the amplitude above ~2px at 30fps.
Why not generate it
Stillness cannot be prompted and neither can a controlled amount of not-stillness. Ask a model for a micro-shake and you get either nothing or a whole different performance. This is the cheapest possible fix for a held frame reading as frozen, and generating it costs a full clip fee for an effect that is four sine waves.
Not run by us. This command is written from the documented filters, not executed here. Check it on a test clip before a client job.
Watch for. Do not generate this. It is a per-frame offset applied locally, and generating it costs a call for something a two-line expression does. Applied over type it also eats the safe area — check the corners of every caption at full offset before mastering.
083Fake dolly zoomdigital zoom against a physical pullfree in post
Use it when
A subject has to stay fixed in frame while the space around it stretches or compresses.
What it should look like
Fake dolly zoom. Movement: the camera travels along its own axis toward or away from the subject while the crop scales the opposite way, so the subject holds its size in frame and the walls behind it change their distance. Speed: the crop ramp is matched to the travel frame for frame; 12 to 20 percent of scale change across two seconds is the band where the background bends and the subject stays sharp. Framing: the subject's head height stays within 3 percent of its opening height for the whole shot — that constancy is the effect. End: settles with the subject the same size it began and a different amount of room behind it.
Do this in your editor — it costs nothing

A keyframe in CapCut, Premiere, DaVinci or any editor does this. The command below is the same move for anyone who prefers a terminal — you do not need one.

ffmpeg -y -i in.mp4 -vf "zoompan=z='1.3333-0.3333*(clip(on/90,0,1)*clip(on/90,0,1)*(3-2*clip(on/90,0,1)))':x='iw/2-(iw/zoom/2)':y='ih/2-(ih/zoom/2)':d=1:s=1080x1920:fps=30,setsar=1" -c:v libx264 -crf 17 -preset medium -pix_fmt yuv420p -c:a copy out.mp4
What to change
1.3333 is the starting zoom and 0.3333 is how much of it is given back across the move — they must stay paired (start Z, and Z-1 as the travel) or the shot will not land back at 1.0. Z=1/(1-A): A=0.25 gives 1.3333/0.3333, A=0.35 gives 1.5385/0.5385, A=0.15 gives 1.1765/0.1765. Bigger A = stronger stretch and a softer opening frame, because the first frame is the most magnified one. 90 = the move length in frames, in all three clip() calls. Swap the sign — z='1+0.3333*S' — to run it the other way, for footage that physically pulls back. HARD REQUIREMENT: the source must already contain real forward or backward camera travel. On a locked-off shot this is only a zoom, and it will read as one.
Why not generate it
The dolly zoom is one move where the expensive half is free. The part that costs money is the physical travel, which the source already has; the counter-zoom is arithmetic. Asking a model for the whole effect spends a full generation on a shot that usually comes back as a plain push with no perspective change at all, because the model has no reason to separate the planes.
Watch for. Do not generate this as a named move. It is built locally: take footage that already carries a straight push or pull, then keyframe the counter-scale against it. Scaling up past about 20 percent softens the subject, so shoot or generate with resolution headroom, and check the head height on the first, middle and last frames — if it drifts, the counter-ramp is off and the effect reads as a wobble.
084Fake whip pantwo pans merged with a blurfree in post
Use it when
Two shots need a fast join and there is no budget for a generated transition.
What it should look like
Fake whip pan. Movement: the outgoing shot slides sideways out of frame and the incoming shot slides in on the same axis, joined at the frame where both are fully smeared. Speed: 4 to 6 frames out, one frame of full blur, 4 to 6 frames in — under half a second end to end. Framing: the two shots share a horizon line and a direction of travel, so the eye lands in the same place on both sides of the join. End: the incoming shot arrives at its own composition with no blur left by the third frame after the join.
Do this in your editor — it costs nothing

A keyframe in CapCut, Premiere, DaVinci or any editor does this. The command below is the same move for anyone who prefers a terminal — you do not need one.

ffmpeg -y -i in.mp4 -i in.mp4 -filter_complex "[0:v]trim=0:4.5,setpts=PTS-STARTPTS,scale=2160:3840:flags=lanczos,crop=1080:1920:(in_w-1080)/2:(in_h-1920)/2[a1];[0:v]trim=4.5:5.0,setpts=PTS-STARTPTS,scale=2160:3840:flags=lanczos,crop=1080:1920:x='(in_w-1080)/2+((in_w-1080)/2)*pow(min(t/0.5,1),2)':y='(in_h-1920)/2',tmix=frames=5[a2];[1:v]trim=0:0.5,setpts=PTS-STARTPTS,scale=2160:3840:flags=lanczos,crop=1080:1920:x='((in_w-1080)/2)*(1-pow(1-min(t/0.5,1),2))':y='(in_h-1920)/2',tmix=frames=5[b1];[1:v]trim=0.5:5.0,setpts=PTS-STARTPTS,scale=2160:3840:flags=lanczos,crop=1080:1920:(in_w-1080)/2:(in_h-1920)/2[b2];[a1][a2][b1][b2]concat=n=4:v=1:a=0,fps=30,setsar=1[v]" -map "[v]" -an -c:v libx264 -crf 17 -preset medium -pix_fmt yuv420p out.mp4
What to change
THE SECOND -i in.mp4 IS THE INCOMING SHOT — point it at your B clip. 4.5 is where the outgoing shot starts whipping and 5.0 is its end; 0.5 is the whip length on each side and appears as both the trim range and the /0.5 inside each pow(). Shorter than 0.25 and there is nothing to see; longer than 0.7 and it stops reading as a whip. pow(x,2) is the acceleration curve — exponent 3 snaps harder out and in. tmix=frames=5 is the smear, and it is where the blur actually comes from, because ffmpeg has no directional motion blur filter — raise it to 8 for a heavier streak. To whip the other direction, swap the x expressions between [a2] and [b1]. scale=2160:3840 doubles a 1080x1920 source to give the crop 1080px of travel; that upscale softens, so generate at 2160x3840 and delete both scale terms if you can.
Why not generate it
This is a transition, which means it has to fall on an exact frame between two specific clips. A model cannot be handed a cut point and it cannot see the shot on the other side of it. The ledger calls this our cheapest cut for a reason: it turns a hard cut into a directed move for nothing, and a client note of 'faster' is a re-render, not another purchase.
Not run by us. This command is written from the documented filters, not executed here. Check it on a test clip before a client job.
Watch for. Do not generate this. It is two finished shots, a directional blur and a slide, built locally — our cheapest cut. The failure is a direction mismatch: if the outgoing shot leaves frame-left and the incoming enters frame-left, the join reads as a mistake instead of a whip.
From our notes
our cheapest cut
085Targeted anchor cropdigital zoom tracking a pointfree in post
Use it when
A detail inside an existing frame has to be brought up without re-generating the shot.
What it should look like
Targeted anchor crop. Movement: the crop rectangle scales down around one named point — a face, a badge, a screen — so that point holds its screen coordinate while everything else travels outward past the edges. Speed: 15 to 25 percent of scale across one to two seconds, eased in over six frames and out over six. Framing: the anchor point moves no more than 10 pixels from where it started, and the crop stops before the source height falls under the delivery height. End: finishes on a tight frame with the anchor on the same third it occupied at the start.
Do this in your editor — it costs nothing

A keyframe in CapCut, Premiere, DaVinci or any editor does this. The command below is the same move for anyone who prefers a terminal — you do not need one.

ffmpeg -y -i in.mp4 -vf "zoompan=z='1+0.4286*(clip((on-45)/75,0,1)*clip((on-45)/75,0,1)*(3-2*clip((on-45)/75,0,1)))':x='(iw-iw/zoom)*(0.5+0.12*(clip((on-45)/75,0,1)*clip((on-45)/75,0,1)*(3-2*clip((on-45)/75,0,1))))':y='(ih-ih/zoom)*(0.5-0.12*(clip((on-45)/75,0,1)*clip((on-45)/75,0,1)*(3-2*clip((on-45)/75,0,1))))':d=1:s=1080x1920:fps=30,setsar=1" -c:v libx264 -crf 17 -preset medium -pix_fmt yuv420p -c:a copy out.mp4
What to change
45 = the hold, in frames, before anything moves (1.5s at 30fps). Set it to 0 to start moving immediately. 75 = how many frames the move then takes. Both appear inside every clip() — change every occurrence. 0.4286 = zoom travel for A=0.30 (1/(1-A)-1). The anchor: x uses 0.5+0.12, y uses 0.5-0.12, which lands on (0.62, 0.38) of the frame — up and to the right of centre. For any anchor (ax, ay) in 0..1, write 0.5+(ax-0.5) and 0.5+(ay-0.5); 0.5 in both is a centred push. To track a point that MOVES, replace the anchor term with a0+(a1-a0)*S, where S is the smoothstep block already in the expression. s= and fps= are the delivery spec.
Why not generate it
No provider will hold still and then start moving — 'hold, then push in' is not expressible in one generation anywhere, so generating it means two clips, two renders and a cut you did not want. And no provider lets you say WHERE to push toward; you describe it and hope. This is the hold-then-move case the zoompan progress expression exists for, and the anchor is a number you type.
Watch for. Do not generate this. It is a crop keyframed around an anchor, done locally. Two things bite: cropping a 1080-high source past about 25 percent puts you under delivery resolution, and small text that looked acceptable wide will not survive the scale — check the legibility at the final crop before committing to it.
086Subconscious push-invery slow dolly during a serious pointprompt
Use it when
A speaker reaches the line the piece exists for and the frame should tighten without anyone noticing.
The phrasing
Subconscious push-in. Movement: the camera advances toward the subject on a straight line along its own axis, closing about a fifth of the distance between them across the shot; the subject holds position in the world. Speed: continuous, around 0.3 pixels of subject growth per frame, with no acceleration and no stop anywhere in the move. Framing: the eyeline stays on the upper third and the headroom shrinks by exactly the amount the face grows. End: finishes in a chest-up frame from a waist-up frame, on the same lens axis it started on.
Watch for. Written as "very slow" or "subtle" this comes back as near-stillness — our measurement is that under-specified action defaults to almost no movement. State the distance closed and the per-frame growth. It is also free in post as a crop ramp, so on a shot that already exists, do not buy a second take for it.
From our notes
or `POST`
087Conversational driftminute breathing movement to keep the shot aliveprompt
Use it when
A talking head is correct and lifeless and needs the frame to breathe under the performance.
The phrasing
Conversational drift. Movement: the camera floats on an irregular path around one position, travelling a few centimetres forward, then left, then back, never returning to the same coordinate twice; the speaker holds position and the background shifts behind them. Speed: the drift covers about 2 percent of frame width every two seconds and changes direction roughly every 40 frames. Framing: the speaker's eyes stay in the upper third throughout, and no drift crosses the shoulder line into a reframe. End: settles anywhere inside the same medium close-up, with the speaker still on the upper third.
Watch for. This is the fix for a dead static talking head, and the way it fails is by being asked for in adjectives — "barely moving", "almost still" — which returns a frozen frame. Give it centimetres and a direction-change interval. If the shot genuinely has to be locked, that goes to the fixed-camera parameter, not to words.
From our notes
the fix for a dead static talking head
088Profile swingpan from straight-on to side profileprompt
Use it when
A face has to change from address to silhouette inside one shot.
The phrasing
Profile swing. Movement: the camera travels a quarter arc around the subject's head, from directly in front to level with the ear, rotating as it goes to keep the head centred; the subject holds position and the background slides across behind. Speed: 90 degrees across two to three seconds, constant, with no hold at the three-quarter angle. Framing: the near eye stays visible until the last 15 degrees, and the nose stays inside the frame as the silhouette turns. End: finishes on a full side profile with the face looking into two thirds of open frame.
Watch for. Models often turn the head instead of moving the camera, which reads as the subject looking away rather than the frame circling them. Write the subject as holding position and the camera as travelling the arc, and check the background: if it has not moved, the camera did not either.
089React revealpan from the speaker to the listener's facerisky
Use it when
A line needs its effect shown on the person hearing it, without a cut.
The phrasing
React reveal. Movement: the camera rotates on its spot away from the speaker and across the room until the listener's face fills the frame; both people hold their positions and only the camera's heading changes. Speed: the rotation covers the distance in about one second, easing out over the final eight frames so the listener's face arrives settled rather than sliding. Framing: the speaker's shoulder stays in frame for the first third of the move so the two read as one space, and the listener's eyes land on the upper third. End: finishes in a single of the listener, looking frame-left toward where the speaker was.
Watch for. Two people in one take is where identity drifts: the listener frequently arrives as a second copy of the speaker, or changes face between the start and end of the rotation. Lock each person to their own reference sheet with exactly one face on it, and compare the listener's face on the arrival frame against the last frame. On a provider that refuses faces in references, this shot does not belong there at all.
From our notes
two identities
090Desk glidelow camera across a workspace showing handsprompt
Use it when
A workspace, a process or a pair of working hands has to be established in one pass.
The phrasing
Desk glide. Movement: the camera travels laterally across the desk surface at about hand height, passing objects that hold their places; the near edge of the desk sweeps through the bottom of frame. Speed: roughly one desk width every three seconds, constant, with no ease at either end. Framing: the hands and whatever they are working stay in the middle third, and the desk edge stays parallel to the bottom of frame. End: finishes with the hands on the far third and one foreground object left soft in the near corner.
Watch for. Count the objects in the prompt — one keyboard, one mug, two hands — because a lateral pass across a cluttered surface is where duplicates appear. Hands are also the part that deforms first: keep them in the middle third rather than letting them exit past a frame edge mid-move.
091Focus shifterfocus between foreground and background speakerrisky
Use it when
Attention has to move between two people at different distances without a cut.
The phrasing
Focus shifter. Movement: the plane of focus travels from the near subject to the far subject while the camera creeps forward by about a hand's width on its own axis; both people hold their positions and neither crosses the frame. Speed: the plane crosses in 12 to 20 frames, slowing through the middle of the travel so the handover is visible rather than instant. Framing: both faces stay in frame the whole time — the near one on one third, the far one on the opposite third — so the viewer can see which is sharp. End: finishes with the far subject sharp and the near one a soft shape holding a third of the frame.
Watch for. Focus control is the weak one. Models commonly render both planes sharp, or move a subject instead of the focus. Build the separation physically in the prompt — real distance between the two, a long lens, a wide stop — and be ready to do the rack locally as a blur ramp across two masks. If the camera must not creep, that is the fixed-camera parameter, not a phrase.
092Diagonal craneup and out simultaneouslyprompt
Use it when
A piece has to leave a subject and hand over the place they are standing in.
The phrasing
Diagonal crane. Movement: the camera rises and retreats at once, climbing about two body heights while travelling back about three, along one straight diagonal through the world. Speed: constant on both axes so the path reads as a single line; the full move runs four to six seconds. Framing: the subject stays on the same vertical third while the ground opens up beneath it, and the horizon enters frame in the final third of the move. End: finishes on a high wide shot with the subject about a sixth of the frame height and ground visible on all four sides.
Watch for. Split into two halves by the model, this becomes a rise then a pull rather than one diagonal. State both distances and say they happen across the same interval. Widening also invents geography — check that the ground revealed at the end matches the ground the shot opened on.
093Orbit-tilt combocircle while tilting down to the handsprompt
Use it when
A shot has to move from who is speaking to what they are doing in one take.
The phrasing
Orbit-tilt combo. Movement: the camera travels an arc around the subject while its heading tilts downward, so the frame leaves the face and arrives on the hands; the subject holds position and only the camera changes place and angle. Speed: 60 to 120 degrees of arc across three seconds, with the tilt spread evenly over that same interval rather than saved for the end. Framing: the forearms stay in frame through the whole arc so the eye has a line to follow from face to hands. End: finishes on a three-quarter view of the hands and what they hold, with the face cropped at the chin.
Watch for. The tilt tends to arrive as a lurch at the end because the two moves are read as consecutive. Write them as one interval. Hands at the end of an arc are where fingers multiply — state the count and the object they hold.
094Panning truckdrive sideways while panning back at an objectprompt
Use it when
A parked or standing subject has to be held while the space around it moves.
The phrasing
Panning truck. Movement: the camera travels sideways past an object that holds its place, rotating back toward it as it goes, so the object stays near the centre while the ground and background stream past the other way. Speed: about one object width of travel per second, with the rotation matched to the travel so the object never drifts more than a quarter frame off centre. Framing: the object stays between the centre and the trailing third, and the background stays readable as a place rather than a smear. End: finishes with the camera past the object, the object on the trailing third, seen three-quarter from behind.
Watch for. Under-specified, the rotation is dropped and this becomes a plain sideways pass with the subject sliding out of frame. Say that the rotation tracks the object and give the tolerance in frame fractions. Named brands and badges on a passing object also trigger refusals on some providers — describe the silhouette, not the marque.
095Pedestal-tilt sweepdrop physically while tilting up to hold framerisky
Use it when
A subject has to gain height and presence without the frame appearing to move.
The phrasing
Pedestal-tilt sweep. Movement: the camera descends vertically through the world from about eye height to about knee height while tilting upward, so the subject keeps the same place in frame as the angle beneath it changes. Speed: the descent runs two to three seconds and the tilt rate is matched to it frame for frame. Framing: the subject's head stays on the upper third throughout while the amount of sky or ceiling behind it grows. End: finishes on a low angle looking up at the subject, head still on the upper third, ground filling the bottom of frame.
Watch for. Models usually deliver one half of this: either the camera drops and the subject slides down the frame, or it tilts up from the same height and the perspective never changes. The test is the relationship between the head and the horizon — if the horizon sits at the same point on the body at the end as at the start, the descent was not generated.
096Zoom-pan swivelzoom out while whipping around a cornerrisky
Use it when
A reveal needs to happen as a jolt at the moment a corner is rounded.
The phrasing
Zoom-pan swivel. Movement: the camera rotates fast around a corner or an edge while the lens widens, so new space enters from the side and from the frame edges at the same time. Speed: 60 to 90 degrees of rotation across about eight frames, with the widening running over those same eight frames and stopping with them. Framing: one vertical edge — the corner, a door frame, a pillar — stays readable throughout as the thing being rounded. End: finishes on a wide frame of the space past the corner, with that edge on the trailing third.
Watch for. Two fast changes inside eight frames is where geometry breaks: walls bend, the corner duplicates itself, and the wide end invents a room that was not in the establishing frame. Generate with a second of headroom either side so the worst frames can be trimmed, and check the wide end against whatever else in the piece shows that space.
097Spiral push-intwist like a screw while pushing inprompt
Use it when
A push needs unease or momentum under it rather than a straight approach.
The phrasing
Spiral push-in. Movement: the camera advances toward the subject while rolling around its own axis, so the horizon rotates as the subject grows; the subject holds its place in the world and only the camera travels and rolls. Speed: 30 to 60 degrees of roll spread across the whole push, with the push closing about a third of the distance in three seconds. Framing: the subject's centre stays within a quarter frame of the middle while the horizon turns, so the roll reads as the camera turning and not the subject tipping. End: finishes in a medium close-up with the horizon off level by the stated amount and the subject centred.
Watch for. Roll is often applied to the subject instead of the camera, which looks like a person leaning rather than a world turning. Anchor the subject's centre and give the roll in degrees. Past about 60 degrees the frame corners run out of image and the edges go soft — cap it or generate wider than delivery.
098Seesaw tiltup, down, up, like unstable footingprompt
Use it when
A shot needs unstable footing or a reeling point of view without moving the camera through the space.
The phrasing
Seesaw tilt. Movement: the camera's heading tilts up, then down past where it began, then up again, each swing shorter than the one before; its position in the world does not change, only its angle. Speed: three swings across two seconds — roughly 20 frames up, 16 down, 12 up — running into each other with no gap. Framing: the subject stays in frame through every swing, sitting on the lower third at the top of the first and the upper third at the bottom of the second. End: finishes level, with the subject on the centre third and the horizon back where it started.
Watch for. Asked for as "unsteady" this returns a generic handheld wobble. Write the three swings, their frame counts and their falling size, in order. A tilt separates nothing at depth, so do not use this where the point is showing how far apart two things are.
099Overhead arccrescent above a flat surfaceprompt
Use it when
A flat lay, a table of product or a process on a surface has to be shown as one continuous pass.
The phrasing
Overhead arc. Movement: the camera travels a crescent path above a flat surface, looking straight down, so the objects on the surface rotate under the frame while keeping their places on the table. Speed: 90 to 120 degrees of arc across four seconds, constant, with the camera height unchanged throughout. Framing: the hero object sits at the centre of the arc and never leaves the middle third, and the surface fills the frame edge to edge for the whole move. End: finishes with the layout turned a quarter from the opening and the hero object still centred.
Watch for. Overhead arcs drift in height, which turns a rotation into a push and breaks the flat-lay read. State that the height is unchanged and check the object scale on the first and last frames. Counts matter here more than anywhere: every item on the surface is named and numbered or it multiplies as the frame turns.
100Blind corner trackround a corner, snapping onto a hidden subjectprompt
Use it when
A piece needs one moment of revelation and the viewer should find the subject rather than be shown it.
The phrasing
Blind corner track. Movement: the camera travels forward along a wall and turns through the corner, and a subject already standing past it comes into frame from behind the edge — the camera moves into the space, the subject does not walk on. Speed: the approach covers about two metres in two seconds, the turn takes 12 frames, and the framing settles across the eight frames after the subject appears. Framing: the wall edge stays on one side of frame through the approach so the viewer knows something is behind it, and the subject arrives on the opposite third. End: finishes on a medium shot of the subject with the corner edge still on the trailing third.
Watch for. The reveal dies when the subject is generated already visible before the turn. Write the subject as present in the world but occluded by the wall until the camera's heading clears the edge, and check frame one: if they are in it, the shot has no reveal. Our own measurement that a single take struggles to cross between two spaces is flagged suspect but not cleared — keep the corner and the room past it as one continuous space, not two.
From our notes
strong REVELATION
101Continuous take loopend exactly where you startedhybrid
Use it when
A piece is built to replay without a visible restart.
The phrasing
Continuous take loop. Movement: the camera travels a closed path through the world — out, around and back — returning to the position and heading it began on, with every object it passes holding its place. Speed: the first second and the last second run at the same rate, because that is where the join is read; a full circuit takes six to twelve seconds. Framing: the opening composition is the target for the closing one — same subject size, same screen position, same horizon height. End: finishes on the opening composition, matched closely enough that the cut back to frame one is not seen.
Watch for. Generate the travel; the join belongs in post. No model returns a final frame matched to its own first frame, so generate two to three seconds past the loop point, then find the closest-matching frame locally and cut there, with a short dissolve if the match is within a few pixels. Exposure and grade also drift across a long take — match the levels on the two ends before looping, or the seam shows as a brightness step even when the geometry lines up.
From our notes
the viral loop mechanic