Timing and exposure
On ones
Also called ones, singles
A new drawing for every frame of the finished film, so 24 drawings for a second of screen time at 24 frames per second. It is the most fluid option and the most expensive, so productions save it for fast action and for camera moves that would judder otherwise.
What it changes in a generated clip. Generated video is on ones by default, because every frame is a fresh sample. That is the mechanical reason so much of it reads as filmed motion wearing a drawn costume. Asking for ones is only worth doing when you want that fluidity on a specific fast beat.
Animated on ones through the fast action, fluid and continuous.
On twos
Also called twos, doubles, shooting on twos
Each drawing is held for 2 frames, giving 12 drawings a second at 24 frames per second. It is the default exposure for most hand drawn 2D. Productions mix it with ones and threes inside a single shot rather than committing to one rate for a whole film.
What it changes in a generated clip. This is the most useful single request in an animation prompt. A repeated frame between every change reads as drawn rather than filmed, and it also hides small inconsistencies, because the eye is given the same image twice before anything moves.
Animated on twos, each drawing held for 2 frames.
On threes and fours
Also called threes, fours, stepped animation
Each drawing held for 3 or 4 frames, so 8 or 6 drawings a second. It reads as choppy on purpose: television animation of the 1960s and 1970s, low budget work, and modern pieces quoting either.
What it changes in a generated clip. It is the strongest lever available against the smooth continuous drift generated video falls into. The trade is that fast movement on fours strobes, so keep the action inside the frame small when you ask for it.
Stepped on threes, deliberately choppy, 8 drawings a second.
Held pose
Also called hold, held drawing, a beat
One drawing stays on screen unchanged for a stretch of time, commonly 6 to 24 frames. Holds carry a great deal of animation. They let a pose land, give an audience time to read an expression, and set up the next move by contrast.
What it changes in a generated clip. This is the highest value thing to ask for in generated animation. Clips drift continuously because nothing told them to stop, and a stated hold gives the shot a shape. Say what holds still, roughly how long for, and what breaks the hold.
The pose holds completely still for a beat, then the head turns.
Moving hold
Also called living hold
A hold that is not quite frozen: the pose keeps a very small continuing movement, such as breathing or a slow settle, so a character can wait without appearing to stop. What qualifies is contested. Some animators mean 2 drawings with a tiny offset between them, others mean any very slow drift, and 3D and 2D departments use the phrase differently.
What it changes in a generated clip. It is the fix for the failures at either end of a generated clip. A completely frozen frame looks like the render stalled, and unrestricted drift looks like video. A moving hold is how a character waits and stays alive.
A moving hold: the pose is held, with only breathing and a slow settle continuing.
Timing versus spacing
Also called spacing
Timing is how many frames a move takes. Spacing is how far the subject travels between one frame and the next inside that time. Two moves can last the same 12 frames and read completely differently, because one accelerates and the other holds a constant speed.
What it changes in a generated clip. Prompts almost always specify duration and never distribution, so a model spreads the movement evenly and the result feels mechanical. Saying where the frames bunch up, at the start, at the end, or around a pose, is what produces weight.
The move takes 12 frames, with the drawings bunched at the end so it slows into the pose.
Easing
Also called slow in and slow out, ease in and ease out, cushioning
Bunching drawings at the start or the end of a move so it accelerates out of a pose or decelerates into one. In drawn animation it is done through spacing rather than through a curve on a slider, but it is the same effect a motion designer means by an ease.
What it changes in a generated clip. Models apply something close to a linear ease to everything, which is the main reason generated movement reads as mushy. Ask per move rather than globally, and say which end, because an ease into a pose with none out of it reads as decisive.
Fast out of the first pose, easing hard into the last 3 frames.
Cycle
Also called loop, walk cycle, repeat
A run of drawings built so the last leads back into the first, then repeated. Walks, runs, flags, flames and machinery are usually cycles, often with the background moving behind a character who stays in place.
What it changes in a generated clip. A generated clip will not loop unless the prompt asks for it, and a loop that nearly matches is worse than one that does not try. Say that the first and last frames are the same pose, and keep the action to one clean repeat rather than one and a half.
A full walk cycle in place, the last frame matching the first exactly so it loops.
Limited animation
Also called limited
A family of methods for making fewer drawings do more work: reused cycles, a held body with only the mouth and eyes moving, sliding whole layers instead of redrawing them, and camera moves across a still. Whether the phrase is neutral or an insult depends on who is using it, since some treat it as a budget compromise and others as a style with its own history.
What it changes in a generated clip. It describes a look generated video is unusually good at, because holding most of the frame still is exactly what stops the drift. It also lets you ask for economy on purpose instead of receiving it by accident.
Limited animation: the body holds, only the mouth and eyes move, the background slides.
Exposure sheet
Also called dope sheet, X-sheet, timing sheet
The frame by frame chart an animation production runs on. One row per frame, with columns for each drawing layer, the dialogue and the camera, so anyone can see which drawing is on screen at frame 37. The names are used interchangeably and vary by studio and country.
What it changes in a generated clip. You cannot hand a model an exposure sheet, but you can borrow its habit of thinking in frame counts rather than adjectives. A prompt written as frame ranges controls a clip far better than one asking for a quick turn.
Frames 1 to 8 hold, frames 9 to 14 the turn, frames 15 to 24 settle and hold.
Staggering
Also called stagger, offset timing, vibration
Two related meanings, and context tells you which. Offsetting the timing of parts so they do not all move together, and a specific shaky effect made by alternating between drawings out of order, used for a shiver, a strain, or a vibrating object.
What it changes in a generated clip. Everything arriving on the same frame is one of the clearest signs a movement was not animated. Asking for the head to lead and the shoulders to follow a few frames later buys more life than any adjective about energy.
The head leads, the shoulders follow 3 frames later, the hands last.
Pose and performance
Key pose
Also called key, extreme
One of the small number of drawings that define a movement, drawn first and usually by the most senior animator on the shot. Everything between them is filled in afterwards. Keys and extremes are not identical in every studio's usage, with extremes meaning the furthest point of a move and keys the storytelling poses, but the words are frequently swapped.
What it changes in a generated clip. Naming 2 or 3 poses gives a generator something to aim at, which is the closest you get to directing a shot rather than describing it. Combined with a supplied first and last frame it is the most reliable control available in generated animation.
Three poses only: crouched low, extended at full stretch, landed and settled.
Breakdown
Also called breakdown drawing, passing position
The drawing between two keys that decides how the movement gets from one to the other. It is not a halfway average: its job is to set the arc and the weight, which is why an animator draws it. Note that the same word means a script or production breakdown elsewhere on a crew, which is an unrelated thing.
What it changes in a generated clip. The middle of a generated move is where a character loses its shape, because nothing specified the route. Describing the midpoint, not only the start and the end, keeps a limb from taking the shortest path straight through the body.
Between the two poses the arm passes wide of the body, elbow leading.
In betweening
Also called tweening, inbetweens, interpolation
Drawing the frames between the keys and the breakdowns. Historically the entry level job on a 2D crew, and now also what software does when it interpolates between two positions.
What it changes in a generated clip. Worth knowing because generated video is not doing this. Frame interpolation, whether in an editing tool or a model filling between 2 supplied stills, averages images and softens whatever it does not understand, which is where melted hands and sliding feet come from. If a move has to be exact, generate fewer distinct poses and hold them rather than asking for a smooth fill.
No smooth interpolation between the poses, hold each pose and change on twos.
Pose to pose and straight ahead
Also called straight ahead action
The two ways of building a shot. Pose to pose sets the key drawings first and fills between them, which keeps timing and structure under control. Straight ahead starts at frame 1 and works forward one drawing at a time, which suits fire, water and unpredictable action but drifts away from the plan.
What it changes in a generated clip. A generator works straight ahead by construction, each moment conditioned on the last, and that is why a clip wanders. Everything that helps, supplied first and last frames, named poses, stated holds, is a way of forcing a pose to pose structure onto a straight ahead process.
Start on the stated pose and end on the stated pose, with the action planned between them.
Anticipation
Also called the wind up
A small movement in the opposite direction just before the main one. A character sinks before jumping, pulls a hand back before throwing, leans away before running in. It tells an audience what is about to happen and makes the main move readable.
What it changes in a generated clip. Actions in generated clips begin the instant the prompt says so, which reads as weightless. A few frames of preparation is the cheapest weight you can add, and it gives the clip a beat structure instead of one continuous move.
A short crouch down as anticipation, then the jump.
Arcs
Also called arcing action, path of action
Almost everything alive moves along a curve rather than a straight line, because it is hinged at joints. Animators check this by tracing the path a hand or a head takes across a run of drawings and looking for a smooth arc.
What it changes in a generated clip. Straight line movement is one of the fastest reads for a synthetic performance, and it is common in generated clips because the shortest route between two states is a straight one. Say that the movement curves, and say which way it curves.
The hand travels in a wide arc, curving up and across rather than straight.
Secondary action
Also called supporting action
A smaller movement running alongside the main one that adds to what it says: a character walking and pulling a coat tighter, or talking and glancing away. It is often confused with follow through and studios differ on where the line falls. The usual distinction is that a secondary action is chosen for meaning while follow through is physics.
What it changes in a generated clip. One action per clip is the safe instruction for generated video, and this is the exception worth making room for, because a single supporting movement is what separates a character from a mannequin. Keep it to one, and keep it small.
She walks steadily, with one secondary action: pulling the coat closed.
Silhouette
Also called silhouette value, readable pose
Whether a pose still reads when filled in as a single black shape. Animators test poses this way because an audience recognises an outline before it registers any interior detail.
What it changes in a generated clip. Generated frames often carry plenty of interior detail that means nothing, with arms lost against a torso and hands merged into a chest. Asking for a clear silhouette targets that directly, and it translates into a drawing better than asking for a good pose.
Clear silhouette, arms held away from the body, hands readable against the background.
Deformation, weight and follow through
Squash and stretch
Deforming a shape as it accelerates and as it lands: a ball flattening on impact and elongating in flight, a face compressing on a hard stop. The convention is that volume stays roughly constant, so a shape that squashes wider also gets shorter.
What it changes in a generated clip. It is the difference between cartoon physics and a rigid object moving quickly. Generated clips keep shapes rigid unless told otherwise, so a bounce reads as something heavy with no give. Say where the squash happens and how far, because unbounded stretch is the failure at the other end.
The ball squashes wide on impact and stretches along its path in the air, volume held.
Follow through and overlapping action
Also called overlapping action
Loose parts keep going after the body stops, and different parts stop at different times. Hair, cloth, ears, tails and anything soft carry on and settle late. Some studios treat these as one principle and others separate them, reserving overlapping action for parts that start and stop out of step.
What it changes in a generated clip. This is where a generated clip most visibly has no physics: coats and hair stop the moment the body does, or float independently for the whole clip. Naming the parts that continue, and roughly how late they settle, works better than asking for realistic cloth.
Follow through on the hair and the coat, settling about 4 frames after the body stops.
Drag
Also called lag
Parts of a figure trail behind the part leading a movement, so a hand trails the arm and the arm trails the shoulder. It is the same underlying idea as overlapping action, described from the point of view of what is behind rather than what is ahead.
What it changes in a generated clip. It gives you a way to ask for connected movement without invoking simulation. Say what leads and everything else has somewhere to be, which prevents the flat result where a whole figure translates as one rigid piece.
The shoulder leads, the arm drags behind it, the hand last of all.
Overshoot and settle
Also called overshoot, settle, cushion in
The pose travels slightly past its target and comes back, usually within a few frames and sometimes with one small second bounce. It is how a movement lands without stopping dead.
What it changes in a generated clip. Generated moves either decelerate into a stop or drift past the mark and never return. An overshoot and settle makes a stop read as a decision rather than as the end of the clip, and it hands off cleanly into a held pose.
The head turns slightly past the mark and settles back over 3 frames, then holds.
Smear
Also called smear frame, stretch frame
One or two heavily distorted drawings bridging a movement too fast to show as a series of poses. A limb becomes a streak, a head an elongated blur of itself. It is on screen for a couple of frames and reads as speed rather than as an error.
What it changes in a generated clip. It is the drawn answer to motion blur, and generated video reaches for the photographic version instead. Asking for a smear frame gets the drawn one, and it is worth pairing with a clean pose either side so the distortion has something to bridge.
A single smear frame bridging the fast turn, with a clean held pose either side.
Multiples
Also called multiple limbs, repeat images
Drawing several copies of a limb or object in one frame to show a very fast repeating movement, such as a waving hand or legs in a run. Related to smears and often used alongside them, though a smear distorts one shape while multiples repeat it.
What it changes in a generated clip. Extra limbs are a well known generation artefact, so ask for multiples only when you actually want them and expect to check the result. If you do not want them, the term gives you something precise to exclude instead of a vague request for correct anatomy.
Multiples on the waving hand for 2 frames, then a single clean drawing.
Impact frame
Also called flash frame, hit frame
One or two frames at the moment of a hit that break the style: a white flash, an inverted colour field, a rough line drawing, radiating streaks. Common in action animation, where the impact itself is often not drawn at all and the frame stands in for it.
What it changes in a generated clip. A model will not produce a deliberate one frame style break unless the prompt puts it there, so hits land softly. It also has a practical use, because a hard frame is a good place to hide a cut between clips.
A single white impact frame on the hit, then the recoil pose.
Technique and medium
Cel shaded 2D
Also called cel shading, toon shading, flat shading
Flat areas of colour with hard edged shadow shapes and visible outlines, the look of paint on acetate cels, although almost nothing has been photographed on cels for decades. Shadow is typically 2 tones, occasionally 3, and not a gradient.
What it changes in a generated clip. This is the technique to state first, because it decides how the rest of the prompt is read. Left unsaid, a model blends drawn and photographic cues into the uncanny middle ground. Stating it also makes any photographic request later in the prompt visibly contradictory, which is useful.
Cel shaded 2D animation, flat fills, hard edged 2 tone shadows, visible outlines.
Painted 2D
Also called painterly animation
Drawn animation with no outlines, soft edges and visible brushwork, closer to a moving painting than to a cartoon. It covers a wide range, from a watercolour background under simple characters to fully painted frames.
What it changes in a generated clip. It is the most useful alternative when outlines are the problem, since models add black outlines as soon as they hear the word animation. Painted styles also tolerate small frame to frame instability, because brushwork is expected to change.
Painted 2D animation, no outlines, soft edges, visible brushwork.
Cutout animation
Also called cutout, paper cutout, rigged 2D
Characters built from separate flat pieces that slide and rotate at hinged joints rather than being redrawn. The word covers both physical paper under a camera and the digital rigs most television 2D is made with now.
What it changes in a generated clip. It suits generated video, because the movement is flat, planar and limited by design, which is the movement a model can hold together across a clip. Name the hinged joints, or you get a fully drawn look with cutout colouring.
Cutout animation, flat pieces hinged at the joints, parts sliding and rotating rather than redrawn.
Stop motion
Also called stopmo, claymation, puppet animation
Physical objects photographed one frame at a time and moved between exposures. Claymation is stop motion in modelling clay, puppet animation with built armatures. The material shows: fingerprints, seams, fibres, and a small unevenness in the movement that no digital process reproduces by accident.
What it changes in a generated clip. Models have a strong association with this look and reach it easily, but they render it too cleanly and too smoothly. The material imperfections and the slight judder are the point, so ask for them, and consider asking for twos, because much stop motion is shot that way.
Stop motion, visible clay fingerprints and seams, slight frame judder, shot on twos.
Rotoscoped
Also called roto, traced animation
Animation traced over live action footage frame by frame, so the motion is human and the drawing is not. Historically done by hand and now also by automated tools, and the word is used for both even though the results differ.
What it changes in a generated clip. It names the thing generated video produces by accident: drawn styling over photographic motion. If that is the look you want, ask for it and you get the coherent version. If it is not, the term tells you what your output has been defaulting to.
Rotoscoped look, drawn outlines and flat fills over naturalistic human motion.
Stylised 3D
Also called stylized 3D, non photorealistic 3D
Computer generated animation that is deliberately not photographic: simplified forms, soft even light, exaggerated proportions and no attempt at a lens. Distinct from photoreal 3D, where camera and optics vocabulary applies normally.
What it changes in a generated clip. It is the one animation technique where some camera language is fair, since 3D productions do use virtual cameras and do sometimes use shallow focus. The distinction still needs stating, because a stylised request next to a named cinema lens drags the result back toward live action.
Stylised 3D render, simplified forms, soft global light, no lens effects.
Effects animation
Also called FX animation
The animation of everything that is not a character: smoke, water, fire, dust, debris, magic, and the way light interacts with all of it. On a traditional crew it is a separate department, because it takes different skills from character work.
What it changes in a generated clip. Effects are usually where a generated animation stops looking drawn, because smoke and water default to photographic simulation even inside a flat scene. Saying the effects are drawn in the same style, on the same exposure rate, keeps the frame consistent.
Drawn effects animation: smoke in flat shapes with hard edges, on the same twos as the characters.
Camera in animation
Parallax
Layers of a flat scene moving across the frame at different speeds, so nearer things travel further, which reads as depth even though nothing in the scene is three dimensional.
What it changes in a generated clip. It is how you get depth in a 2D scene without inviting the photographic depth cues that break the style. Say how many layers and which direction, because given only the word, a model often produces a general slow drift rather than separated layers.
Horizontal pan with 3 layers of parallax: foreground fast, midground slower, sky nearly still.
Multiplane
Also called multiplane camera, layered setup
The rig that produced parallax in traditional animation, holding painted layers at separate distances under one camera so each could be moved independently. The word now describes the layered approach as much as the machine.
What it changes in a generated clip. Thinking in stacked layers rather than in a 3D space keeps a move flat and planar, which is what generated animation can hold together. Describing a scene as named layers also gives you something to lock, and a locked sky is one fewer thing that can drift.
Multiplane setup: painted sky layer locked, midground trees sliding, foreground grass fastest.
Truck in
Also called push in, scale up on the artwork
The animation camera moving toward the artwork, which on flat art means the drawing is enlarged rather than the perspective changing. Foreground and background scale by the same amount and nothing moves around anything else, which is why it looks unlike a live action dolly.
What it changes in a generated clip. Ask for a dolly in on a flat scene and a model may add perspective shift and parallax a cel setup could not produce, which reads as a 3D scene wearing drawn textures. Naming a truck in, or simply a scale up on the artwork, keeps the move honest to the technique.
Slow truck in on the held drawing, a straight scale up with no perspective change.
Background pan
Also called pan, sliding background, pan cycle
The background artwork slides horizontally behind a character who often stays in place, walking or running on the spot. Traditionally the background is painted long and wide and moved past the camera in even increments.
What it changes in a generated clip. It is the most reliable camera move in generated animation, because the character holds a cycle while only the background changes and there is little for a model to reinvent. Say the direction and that the character stays centred, or the framing wanders.
The background pans right to left at a constant speed, the character stays centred, walking in place.
Camera shake
Also called shake, impact shake
A short violent displacement of the whole frame at a moment of impact, usually 2 to 6 frames and then completely still. In animation it is drawn or applied as a move on the artwork rather than captured, so it starts and stops exactly on cue.
What it changes in a generated clip. Models read any shake request as continuous handheld motion, which is a live action idea and does not belong in most animation. Bounding the shake in frames, and saying the frame is locked otherwise, gets the drawn version.
The frame is locked throughout, with a 3 frame hard shake on the impact only.
Line and fill
Outline colour
Also called line colour, self coloured line
The colour of the drawn line around a shape. Black is the exception rather than the rule in professional work. Productions commonly use a dark brown, a darker shade of the fill the line encloses, a colour that changes across a character, or no line at all.
What it changes in a generated clip. A model defaults to black outlines, and black outlines on everything are the fastest way for a piece to read as generated. Naming the line colour takes one clause and changes the register of the whole image.
Dark brown outlines, no black line anywhere, shadow shapes outlined in a darker shade of their own fill.
Line weight
Also called line width, tapering line
How thick the drawn line is and whether it varies. A line can be even in width, taper at its ends, or run heavier on the shadowed side and around the outer silhouette. Characters are commonly outlined more heavily than backgrounds.
What it changes in a generated clip. Line weight is the detail most likely to change between shots in a sequence, which breaks continuity even when everything else matches. Write the treatment once and repeat it in exactly the same words in every shot rather than paraphrasing it.
Even weight outlines on the character, thinner lines in the background, no tapering.
Line boil
Also called boiling line, wobble
The visible shifting of a hand drawn line from frame to frame, because each frame was drawn separately and no two are identical. Some productions treat it as an artefact to remove and others as a signature of hand work and keep it deliberately.
What it changes in a generated clip. Generated animation produces this constantly and involuntarily, which is why so much of it looks like it is simmering. Naming it lets you take a side: ask for a clean stable line when you want control, or ask for boil when the hand drawn signature is the point.
Clean stable line, no line boil, outlines consistent from frame to frame.
Shading steps
Also called tone count, two tone cel, cel shadow
How many distinct values a surface is broken into. A 2 tone treatment is a base colour plus one shadow, 3 tone adds a highlight or a deeper shadow, and a soft gradient means no steps at all. This one choice moves a drawing across decades of animation history.
What it changes in a generated clip. Left unstated, a model produces a soft gradient, because that is what most images in the world look like, and a gradient on a flat character is the clearest hybrid tell there is. Give a number, and say whether the shadow edges are hard.
2 tone shading only, hard edged shadow shapes, no gradients.
Motion lines
Also called speed lines, action lines
Drawn streaks trailing a moving object or filling the background behind it, standing in for speed. A drawn convention rather than an optical effect, and unrelated to what a camera does at a slow shutter speed.
What it changes in a generated clip. It is the substitute for motion blur inside a drawn style and it has to be asked for, because a model supplies blur instead. Say whether the lines trail the object or fill the background, since those two read very differently.
Drawn speed lines trailing the fist, no photographic motion blur.
Cel over painted background
Also called character and background split
The very common arrangement where the background is painted in one technique and the characters are drawn and filled in another, then layered over it. The background is usually softer, more detailed and less outlined than anything moving in front of it.
What it changes in a generated clip. A model renders a whole frame in one technique unless the prompt splits it, which produces either a character that sinks into its background or a background as graphic as the character. State both treatments separately, and say which one carries the detail.
Painted watercolour background with no outlines, cel shaded characters with flat fills layered over it.
Common questions
What does animating on twos mean?
Each drawing is held on screen for 2 frames instead of 1, so a second of film at 24 frames per second contains 12 drawings rather than 24. It is the standard exposure for most hand drawn 2D animation. In a prompt it is one of the most useful things you can ask for, because the repeated frame between changes is a large part of what makes motion read as drawn rather than filmed.
Why does my AI animation look like video and not animation?
Because a generated clip is on ones by default, with every frame a fresh sample and nothing holding still. That produces continuous movement, which is filmed motion, not drawn motion. Two requests fix most of it: an exposure rate, usually on twos, and at least one held pose where a single drawing stays unchanged for a beat. Removing photographic requests such as depth of field, motion blur and handheld camera helps as much again.
What is a held pose in animation?
One drawing left on screen unchanged for a stretch of time, commonly 6 to 24 frames. Holds do real work: they let a pose land, give an audience time to read an expression, and make the next movement register by contrast. They are the thing generated animation most obviously lacks, because nothing tells a model to stop moving.
What is squash and stretch?
Deforming a shape as it accelerates and as it lands, so a ball flattens on impact and elongates in flight. The convention is that volume stays roughly constant, so anything that squashes wider also gets shorter. It is the main source of weight in cartoon movement, and generated clips leave it out unless the prompt asks for it, which is why bounces in them read as rigid.
What is the difference between timing and spacing?
Timing is how many frames a movement takes. Spacing is how far the subject travels between one frame and the next inside that span. Two moves of the same 12 frames can read completely differently depending on whether the frames are spread evenly or bunched at one end. Prompts usually give a duration and never a distribution, which is why generated movement often feels mechanical.
Should I use a lens or a focal length in an animation prompt?
Not for flat 2D, cutout or stop motion work. Naming a lens and an aperture on a drawn style pulls photographic cues into the image and produces a hybrid that reads as neither. The exception is stylised or photoreal 3D, where productions do use virtual cameras, and even then a named cinema optic will drag the result toward live action.
What is limited animation?
A group of methods for making fewer drawings do more work: reused cycles, a held body with only the mouth and eyes moving, sliding whole layers instead of redrawing them, and camera moves across a still. Whether the phrase is a criticism or a description of a style depends on who is using it. For generated video it is worth knowing, because holding most of the frame still is what stops the drift.
What is a smear frame?
One or two heavily distorted drawings bridging a movement too fast to show as ordinary poses, so a limb becomes a streak for a couple of frames. It is the drawn equivalent of motion blur, and it has to be requested, because a model will supply photographic blur instead. Ask for a clean pose either side of it, so the distortion has something to bridge.
How do I keep an animated character consistent across shots?
Repeat the character description in identical words in every shot, including height, proportion, hair, wardrobe and palette, and never paraphrase it. Anchor the eye construction, which is the strongest identity cue, and pin the line treatment, meaning line colour, line weight and the number of shading steps, in the same words each time. Keep the number of characters in a shot low, because each additional one multiplies the drift.
What does it mean to animate on threes?
Each drawing is held for 3 frames, so 8 drawings a second at 24 frames per second. It looks choppier on purpose and is associated with television animation of the 1960s and 1970s and with work quoting that era. It is a strong tool against continuous drift in generated video, with one caveat: fast movement held that long strobes, so keep the action small when you ask for it.
