Motion Design Judgment Methods

What This Skill Is

It documents recurring practices found in a body of reference works, along with four questions for each practice:

What is actually seen in the frame (with work ID, timecode, measurement method).
Why it might be organized this way—in relation to content, visual hierarchy, and rhythm. This is interpretation, not the author's statement.
Under what circumstances it is applicable and where it is not.
What trade-offs are involved and how to verify in your own draft.

It is not a style template. Pasting together the rulers, portholes, brackets, inverse flashes, and checkerboards from the source material produces a derivative work. What can be taken away is the job each element performs in its original context, along with a method for determining "does my piece have this job to do?"

Three Bottom Lines

These three principles determine whether this skill is useful—more important than any specific judgment below.

One: Separate observation, interpretation, and hypotheses. What can be determined from a completed piece is only what is seen and measured. Software, project structure, and the author's intention cannot be inferred from the final output. When giving advice, make it clear which statements are factual observations from the source material and which are your inferences. This lets the user know what they can trust directly and what they need to judge for themselves. With only a still frame, even less can be determined: how long it lasted, how it arrived, how it left—still frames cannot tell you. When evidence is insufficient to choose between two interpretations, present both and state plainly "cannot be determined."

Two: Borrow the job, not the accessory. The same job (preventing the eye from needing to reorient during fast cuts, preventing high-frequency content from scattering, catching an accent within a single frame...) is performed by different elements in different works. First list the jobs your piece needs to do, then find elements from its own content to perform those jobs. A complete reference table is in references/structure-and-motif.md section 7.

Three: Values belong only to the segment in which they were measured. "0.9 decay per frame" or "large type occupies 18% of frame height" are measurements from a specific segment of a specific work. They can help you determine order of magnitude or serve as a starting point, but are not to be taken as rules. Always cite the source when referencing numerical values. Statements involving frame counts must also indicate the level of evidence: frame-by-frame analysis, full-scene statistics, every-other-frame or single-frame analysis, or inference from still frames or other sources. Only the first level allows the phrase "X frames." See references/analysis-method.md section 7.

Additional notes:

B01—B14 are 14 reference images. 5 of them correspond to a single frame of V05 or V00 and are not independent works, nor are they assumed to be screen captures from the final video; the provenance of the remaining 9 has not been confirmed. See references/cases.md section 4.

First Determine the Task Type

What the user wants to do	How to proceed	Where to read
Design a new piece, cover, or poster	Follow the "Design Sequence" below	Read the corresponding reference file for each judgment used
Critique or revise an existing draft	Follow the "Critique Sequence" below	references/rhythm-and-sound.md section 10, and the "How to Check" sections at the end of each file
Ask a specific question (whether a cut point should land on a beat, how large type should be, how long a line of text should hold, whether to add motion blur)	Find the corresponding item among the "Twenty-Five Judgments," read the section it points to, then answer following the "Response Format"	The corresponding reference file
Select or handle title type (weight, width, tracking, multiple typefaces, stretching, cropping, masking, repetition)	First determine the job of the type, then its appearance; do not identify font names	references/typography.md
Arrange a single frame's composition (center of gravity, white space, symmetry vs. offset, layers, use of geometric elements)	First measure, then look; each geometric element must have a stated job	references/composition.md
Work with flashes, inversions, empty frames, strobing	First determine the relationship between this frame and its surrounding frames, its position, and what remains static; check the frequency per second	references/single-frames.md
Set easing and keyframe curves	First determine the intended texture, then select the shape of the arrival; what is measured is a sequence of positions, not parameters	references/motion-and-transitions.md section 10
Deconstruct a reference video	Measure and observe following the analysis method sequence; write notes in its format	references/analysis-method.md
Only have still images (posters, screenshots, author's preview images) and want to know what the final video looks like, or want to replicate its motion	Describe only what is in the still; clearly state what still frames cannot reveal; if a finished video exists, first match the still to its frame, then examine frame by frame	references/stills-and-motion.md
See interface symbols in the frame (checkerboard, popup, code) and want to know what they are or whether to use them	First look at its boundary, what it follows, and when it appears; if its origin cannot be determined, state that honestly	references/color-and-texture.md section 9
Replicate a segment from reference	First determine the camera, then list elements, then compare version by version	references/replication.md, and the analysis method
Want to make something "like [that]"	First break "that" down into jobs, then find elements from the user's own content to perform those jobs	references/structure-and-motif.md section 7, references/cases.md

Reference files are independent; only read what is needed. Each file begins with a table of contents.

Design Sequence

Start from content and narrow down step by step. The first four steps do not involve any specific graphics.

Write down one thing the audience should take away from this piece. A word, an image, a sense of speed, a lyric. This determines who is the subject; everything else yields to it. Subjects vary greatly in the source material: text and graphics (V01—V05), illustrations (lyric videos), 3D renders (V22—V25), narrative (V34). Different subjects lead to different answers at every subsequent step.
Separate text into what needs to be read and what only needs to be seen. What needs to be read requires time, medium-to-large size, stable position, and clear edges; what only needs to be seen can be very large or very small, and can be distorted. After this step, type size, lyric position, and which words can be deformed are largely determined. Type size and hold time are determined together: the more text to be read, the longer the time. See references/layout-and-type.md sections 1—5.
Layer the music and assign a type of visual change to each layer. Sections, measures, accents, and timbre density are each picked up by visual changes of different magnitudes. Do not let the same type of change serve two layers simultaneously, or the larger layer will have nothing to mark it. See references/rhythm-and-sound.md section 1.
Determine what remains static. A running motif, an anchor point across cuts, a container for content, an axis linking elements. Establish what stays still first, so moving elements have a reference. These must come from the piece's own content. See references/structure-and-motif.md sections 1, 2, 7 and references/motion-and-transitions.md sections 3, 8.
Set a budget. How many accent colors and what each controls; how many large type instances; roughly how many changes per second; how much material can be prepared. The tighter the budget, the more each change feels like a single event. See references/color-and-texture.md sections 1—3.
Determine the texture family and motion texture. Is the frame "data and print," "seen through a lens," or "recorded"; is motion abrupt, continuous spatial, or slow drift. Transitions and entrances follow the same family. See references/color-and-texture.md section 7 and references/motion-and-transitions.md sections 1, 5, 6.
Plot a density curve. Each segment: sparse, dense, or empty. Intensity is perceived through contrast, so decide first where it will be emptiest. See references/rhythm-and-sound.md section 8.
Layout a few key frames. For each frame, determine four things: where the center of gravity is and to whom the empty side belongs; how many layers and what separates each two; which line or block serves as the baseline, who follows it and who breaks it; the job and appearance of each block of type (bold type as graphic, fine type needs space and time). Each geometric element must have a stated job. See references/composition.md and references/typography.md.
For important elements, assign three numbers and one shape: entrance duration in frames, full hold duration in frames, exit duration, and the shape of the arrival. For exits, first check if the next change (a cut, an in-place replacement, a camera move) can be borrowed. The shape of the arrival follows the texture: arrive in one frame, stepwise jumps, fast-then-slow stop, constant speed. For very dense full-frame compositions, think in two layers: the one or two large shapes that will be seen during playback, and the details visible only on pause. For single-frame flashes or empty frames, first determine whether they serve as emphasis, transition, or preview, and count the number of flashes per second. See references/motion-and-transitions.md sections 9, 10, references/single-frames.md, and references/stills-and-motion.md section 6.
After creating a draft, measure it and compare with the plan. Arrangements like "cuts follow the measure" or "accents are caught by on-screen events" can be measured on the rendered draft. See "Critique Sequence."

When providing a plan to the user, write out the decision and rationale for each of these ten steps in order. If the user has not provided music, content, or the context of release, either ask first or state what you have assumed.

Twenty-Five Judgments

Each judgment is stated as a conclusion. The supporting evidence, applicable conditions, and trade-offs are in the referenced sections. Read the relevant section before answering a specific question.

Rhythm (references/rhythm-and-sound.md)

On what level of beat do cuts land? Among the source material, 9 pieces show significant locking; cuts are locked to two beats, three beats, four beats, or eight beats; only 1 piece locks to one beat; none lock to half-beats. In time, all are over half a second. Interpretation: Cuts require the viewer to re-read the screen. Faster cuts exist, but they follow fixed frame counts and function as a texture. (Section 2)
Who catches the accent? In six cases (V05, V04, V08, V07, V17, V03), the accent lands on an on-screen event (an in-place transformation, a weight change, a color change, a protruding edge); these events are 1.9—3.2 times random chance. Cuts show no preference for accents. Interpretation: Accents need a change that completes within one frame and does not require the eye to find a new subject. A composition stable over several beats is needed. (Section 3)
How much motion per second follows what? In 21 out of 23 pieces, motion quantity correlates positively with high-frequency energy; correlation with overall loudness is lower. Interpretation: The overall loudness of mastered music is flat; high-frequency energy is closer to the perceived sense of "fullness." There are also counter-examples (V24). (Section 4)
What marks a section change? Background color, frame size, lyric arrangement, or the entire color palette. Reserve the largest change for section boundaries. (Section 5)
Should you use a fixed step rate? Rates of 5, 8, 10, 12, 15 steps per second have been measured, but they are applied only to one layer or one segment, not locked to the beat. It is a material texture for that layer, not the rhythm. Items needing precise timing should not be placed on a stepped layer. Measure step rate only from frame-by-frame data; every-other-frame sampling can turn continuous motion into stepping. (Section 6)

Motion and Transitions (references/motion-and-transitions.md)

Where is speed placed within a shot? In six measured shots with cuts, large motion is concentrated near cut points: four shots have large motion immediately after the cut-in, two shots have it in the last few frames before a cut-out; the middle is slow drift, constant speed, or static. Interpretation: The cut masks the large displacement, saving the middle for viewing. Not applicable when shots are very short. (Section 2)
What remains unchanged during a cut? In graphic-style pieces, most cut points have at least one unchanged element: a centered graphic, a framework layout, a lyric layer, a 2D marker, or the entire scene arrangement. Interpretation: The eye does not need to reorient, making high-frequency switching readable. May be omitted at section boundaries. (Section 3)
Which layer should be blurred? Flat graphics remain sharp at high speed; 3D renders carry blur over which 2D annotations remain sharp. Interpretation: Blur says "passing through," sharpness says "read me." Ask of each layer whether it needs to be read during motion. (Section 7)
How do new elements enter? Hard pop-in, silhouette-then-full shape, blur-to-sharp, character-by-character, staged assembly (skeleton, content, color)—almost no fades. The entrance method is a material property of this world, following the texture family; what arrives first stays longest; the order defines hierarchy. (Section 6)
How long does an element hold and how does it exit? Among 14 frame-by-frame examined instances, only 3 had a dedicated exit animation; the rest borrowed a cut, in-place replacement, or camera move; a one-frame marker often precedes the replacement. Hold time follows the amount of reading: a large word filling the frame holds 5—10 frames, a lyric line to be read holds 0.9+ seconds, a full block of small type holds only 2—9 frames. Interpretation: Entrances deserve a distinct form; exits are given to the next change. Small sample, mostly text. (Section 9)
What shape does arrival take? Among over a dozen frame-by-frame measured instances, these shapes appear: arrive in one frame; stepwise jumps (two or three positions, no in-between frames); appear at max size then shrink; fast-then-slow, remaining amount multiplied by approximately 0.6—0.9 per frame; constant speed; accelerate before cut-out. No overshoot, no slow-fast-slow symmetrical easing has been measured. The shape follows the texture. What is measured is a position sequence; what keyframes or parameters correspond cannot be determined from the final video. (Section 10)
What goes into one-frame flashes, inversions, and empty frames, and where are they placed? Most examined instances share the same layout or layer with the surrounding frames; only polarity, background color, or "has content" changes. Placed within a shot, they emphasize; placed at a cut, they add a threshold to the transition; placed 2—3 frames before a cut, they preview. During a flash, something remains static. In most pieces, they do not land on musical accents but cluster with cuts at a specific part of the measure. More than three full-frame flashes per second carries a photosensitivity risk; two instances in the source material exceed this. (references/single-frames.md)

Layout and Typography (references/layout-and-type.md)

How large is type, and how many size tiers are used? Large type appears 1—3 times, occupying 10%—45% of frame height; most text is under 2.7%; the middle tier is rare, used only when lyrics need to be read. The largest is not necessarily the primary: large type that is darkened, blurred, or matches the background color functions as background. Small type must be checked at delivery resolution and bitrate. (Sections 1—3)
Where does the lyric sit? As a subject, or as a subtitle. Depends on who is the lead in the frame. When it is the subject, it is an independent design job. Sung lyrics appear at irregular intervals and rarely exit independently. (Section 5)
What frame contains the image, and does the frame size change? A window separates images of different styles or graphic layers; changing frames allows limited illustrations to appear in more ways; multi-frame adaptation reflows by "follow the edge, follow the subject, can be omitted." (Sections 7, 8)
Are typefaces thick or thin, wide or narrow, and how are multiple fonts combined? Type that occupies over 10% of the frame and needs to be seen first is mostly thick sans-serif. Thin type appears in very small labels or in very empty, long-hold frames. Tight tracking and leading combine multiple lines into one pliable block; loose tracking turns a word into a line. When two typefaces are combined, the difference is large; the distinctive one is used only on the largest element. Even large, thick type can be reduced to a background texture by lowering contrast. Typeface names cannot be determined from the final video. (references/typography.md sections 1—4)
To what extent can type be enlarged, deformed, cropped, masked, or repeated? Depends on whether the word still needs to be read: if yes, keep one instance legible; if no, can be pushed to the extreme. Cropped type says "it's bigger than the frame"; masked type creates depth; repetition offers two readings of the same information. The appearance of the type also determines how it enters: blocky type jumps in, thin type gathers, falls, or resolves from dots. (Same, sections 5—9)

Composition (references/composition.md)

Where is the center of gravity, and how much white space? In 24 measured frames, 14 have a center of gravity within 3% of the frame center horizontally. When offset, the empty side is given to the subject, to something about to enter, or for breathing room, and often a single element bridges both sides. Interpretation: Centering is least demanding on the eye during fast cuts. Symmetrical compositions give change to time. (Sections 1—4)
How many layers, separated by what, and who aligns or offsets? Most frames have three to four layers, separated by scale, brightness, sharpness, a unique color, or occlusion; the largest element is often in the back. Establish one baseline (a line, an edge, a centerline); most elements follow it, one or two break it. Compositions relying solely on brightness for layering are vulnerable to inversion. (Sections 5, 6)
Should lines, dots, circles, stars, arrows, or grids be used? Only if each can state its job: a line is a baseline or track, a dot counts, a star is a one-time impact, a ring pulls the eye back to the subject, parentheses and quotation marks mark and frame. Three readings of a grid: white with light gray reads as "transparent," black with white is a pattern, a large grid is a ground plane. (Sections 7, 8)

Color, Texture, and Space

How many colors are used, and how are they used? Most works use less than 40% of the frame in color: achromatic, one accent color, a pair that swaps, or a palette per section. The fewer the colors, the more each change feels like a single event. When swapping a color pair, check the brightness difference. (references/color-and-texture.md sections 1—3)
How are background color and texture determined? Leave a margin from pure white or pure black so that white flashes and highlights have room. Texture declares through what the frame is being seen and unifies disparate origins. (Same, sections 4, 7)
How are 2D and 3D placed together? Either make the 3D follow 2D rules (flat color, two brightness levels, hard edges, no blur) or clearly layer the 2D on top of the 3D (sharp, unlit, not moving with camera). Do not stop in between. (references/space-2d-3d.md sections 2, 3)

Stills and Symbols

What can a still frame tell you? Among five matched author preview images, in the final video they all hold for only 1—9 frames, each being the most complete moment of its shot. A still frame cannot tell you five things: how long it lasts, the order in which elements appear, whether there are flashes before or after, what the camera is doing, and which layer is static. What a still gives more than a platform-compressed video is the color of small type and fine lines. When only a still exists, describe only the still. To discuss motion, first match it to the finished video. Very dense full-frame compositions appear as a single "dense" flash during playback; details are for those who pause—this is an interpretation. (references/stills-and-motion.md)
What do interface symbols (checkerboard, popup, code) do in the frame? Look at three things in order: what shape is its boundary, what does it follow (world, object, screen), and when does it appear. In five pieces, the checkerboard is white with approximately 80% brightness gray; it appears in four instances at moments when "something has been removed." But some are attached to the world, others fixed on the screen. What it does in the frame can be described; whether it was intentionally placed or came from another source cannot be determined in the screen-fixed cases. (references/color-and-texture.md section 9)

Critique Sequence

First, determine the intent. Ask the user, or infer from the draft: what visual change was assigned to each musical layer, what is the anchor point, which type needs to be read, what each accent color controls. Drafts without a clear intent usually have problems here; first fill it in following the "Design Sequence."
Measure. Treat the rendered draft as source material. Measure it using the same scripts used for analyzing reference clips (see "Scripts" below).
Compare.

Intent	What to look at	What counts as matched
Cuts follow the measure	`audit.py` Row B: periodicity R values	The intended period is strongest and significant
Accents caught by on-screen events	Row A: pulse lift and z	Lift clearly > 1, z > 2
Motion quantity follows timbre density	Row D: `motion_high`	Clearly positive (negative if intentionally reversed)
Section change changes the world	Row E; `vis.py colorbar`	Hit count clearly above random chance
Cut point preserves an anchor point	`vis2.py cuts`	Each cut has an identifiable unchanged element
Layer to be read stays sharp at high speed	`vis2.py top`	No trailing edge at original resolution
Type size tiers	`typesize.py`	Number of large type instances matches intent
Small type still readable	Encode at delivery specs, then check frame by frame	Smallest lines still legible or at least form rows; fine type differentiated by color still has color
To-be-read type has enough reading time	Step-1 strip; count frames of full hold	Consistent with magnitude in references/motion-and-transitions.md section 9, or explain why different
Entrance, hold, exit	Same, for each important element fill four columns	Each element can describe how it arrives, holds for X frames, and how it leaves; in-place replacement not read as jitter
What a dense full-frame composition leaves in playback	Watch at normal speed once, write what you remember; then pause and compare	What is remembered is the one or two things intended to be seen
Type weight, width, and difference between two typefaces	`typemetrics.py` to measure weight, width, tracking, leading; shrink the frame to see	Type-as-graphic remains a solid shape when shrunk; two typefaces are distinguishable
Center of gravity, white space, offset	`compose.py`	Matches intent; for offset frames, state who the empty side belongs to
How many single-frame flashes, per second	`oneframe.py`, then `ofsheet.py` to examine each	Each identifiable as emphasis, transition, or preview; no more than three full-frame flashes in any one second
Shape of arrival	`elemtrack.py` for per-frame position and "ratio of remaining distance"	Matches intended shape (arrive in one frame, stepwise, fast-then-slow, constant speed)

Give feedback. For each item, state clearly: what it currently is (measured or observed), where it differs from intent, how to fix it, and what trade-off the fix introduces. Do not use numerical values from the source material as a passing grade. No single piece in the source material scores high on every measure.

When scripts cannot be run (only screenshots, only text description), check the same table item by item using visual inspection, and state that it is an estimate. With only screenshots, any time-dependent item (cuts, accents, motion quantity, holds, entrances/exits) cannot be checked; state this clearly and do not guess from the screenshot.

Response Format

A piece of advice should contain five elements:

What to do.
Why—label this as interpretation.
Basis—which works, at what timecode, and how measured.
When not to do this.
How to verify.

Example

User asks: The chorus needs more punch. Should I cut on every beat?

Bad answer: "Reference a certain fixed style; cut every 3 frames and add inverse flash frames." It gives an unconditional numerical value and treats an accessory as a method.

Good answer skeleton:

Suggestion: Keep cut points on every two beats or on the measure. Assign each beat to an on-screen event: in-place transformation, background color change, elements popping from edges.
Why (interpretation): After a cut, the viewer needs to re-read the screen. Changes that complete in one frame and keep the skeleton static do not require this.
Basis: Among 9 pieces with significant locking, only V20 locks cuts to single beats, each beat lasting half a second. In V04 and V05, on-screen events catch the accent at 2.7 and 3.2 times random chance; cuts show no preference.
When not to do this: When the piece is slow enough that a beat is over half a second, cutting on every beat can still be readable. When no on-screen element can change in place (all live-action or 3D render), the source material uses cuts to catch accents (V21, V23, V24).
Trade-off: Requires a composition stable over several beats. Continuous full-frame inversion carries a photosensitivity risk; frequency and area must be controlled.
Verification: After rendering, run audit.py and check if pulse lift is clearly greater than 1.

Habits in language:

Say "seen in V04, V16"—do not present a partial observation as a universal rule. Case details are in references/cases.md.
Say "likely because"—do not say "in order to."
When reporting frame counts, indicate how they were measured. For frame-by-frame, give frame numbers. For every-other-frame or single-frame, say "it exists" but not the exact count.
When referring to images, say "the content, composition, and scale match a specific frame"—do not say "it is a screen capture from the final video" unless there is other evidence.
When the user wants to make something "like [that]," first describe the jobs performed by that work, then ask the user what elements from their own content can perform those jobs.

Things Not to Do

Do not transfer a specific set of elements from the source material wholesale into a new design.
Do not turn a measured numerical value from one segment into a rule or the habit of an author.
Do not infer software, plugins, or project structure. UI symbols and code appearing in the frame are design elements, not production records.
Do not infer motion, production methods, or author intent from still frames. When only stills exist, describe only what is in the still and clearly state what it cannot reveal.
Do not write frame counts based on every-other-frame strips. Frame count statements must have frame-by-frame evidence.
Do not count an author's published images as independent works, nor assume they are screen captures from a final video. When provenance is not confirmed, keep it as uncertain.
Do not definitively assign an origin to a symbol (e.g., a transparent checkerboard) when evidence is insufficient. State its appearance, position, and timing in the frame.
Do not identify font names. Type in the final video has been scaled, compressed, and processed; describe its appearance (weight, width, serif or sans). To find a name, consult production notes and release info; if not found, write "unconfirmed."
Do not write a measured position sequence as the author's keyframes, curve parameters, or expressions. You can say "a possible approach" and label it as speculation.
Do not write an inferred order as the author's creative process. "Skeleton first, then content" is the order elements appear in the final video, not the author's working order.
Do not present any specific font, flash, or curve as mandatory. Every item has its applicable conditions.
Do not transcribe lyrics; only discuss their arrangement.
Do not recommend large-area, high-contrast, high-frequency full-screen flashing without warning of the risk.
Do not modify the user's existing project, final video, or documents unless requested; save experiments and corrections separately.

Scripts

scripts/ contains the analysis scripts. The same set is used for deconstructing reference clips and measuring your own draft. Requires ffmpeg, ffprobe, and Python with numpy, Pillow, opencv-python.

python -I scripts/scan_all.py <video_path> <ID> [bpm]

If ffmpeg is not in the system path, set the environment variable FFMPEG_DIR. Intermediate results are written by default to motion_work/ in the current directory; change with MOTION_WORK.
Speed must be manually verified after being provided: first look at the candidates output by report.py, then use acf.py to exclude harmonic relationships.
The function of each script, the definition of each metric, and the actual errors this method has made are in references/analysis-method.md. Before reading any number, at least read the section "Actual Errors Made" in that document: hard cut count is often inflated, every-other-frame sampling can create both rhythm and duration, thumbnails cannot judge blur, color measured from JPEG is biased, looking similar does not mean it is the same frame—each of these has previously overturned a conclusion.
Six scripts added in the third round for frame-by-frame verification: still_match.py matches an image to a frame in the video; local_events.py finds frames where small elements appear or disappear; domcolor_seq.py gives per-frame dominant color; onset_at.py aligns onsets with frames; find_checker.py and chk_phase.py find checkerboards and determine what they follow.
Six scripts added in the fourth round: oneframe.py finds frames that hold for only one frame and gives a rough classification; ofsheet.py arranges these frames together with surrounding frames into a comparison sheet; typemetrics.py measures the weight, width, tracking, and leading of a block of type; compose.py measures the center of gravity, white space, and weight of left/right/top/bottom of a frame; elemtrack.py tracks the position and size of an element frame by frame; mgrid.py composes frames from different videos into a single image. These all provide candidates and rough numbers; each should be reviewed against the actual frames before use.

Reference Files

File	Content	When to read
references/rhythm-and-sound.md	Music layering; cut points, accents, motion quantity, sections, fixed step rate, beat length vs. framerate, density	Any music-related decision; critique comparisons
references/motion-and-transitions.md	Motion texture, speed curves, anchor points, background swap & inversion, transitions, entrances, motion blur, frames, hold & exit, shape of arrival	Designing shots, transitions, entrances and exits; determining how long each thing holds and what shape it arrives in
references/single-frames.md	Frames that hold for one frame: how many, what is in them, where they are placed, sequences, what stays static, relationship to music, playback vs. pause, photosensitivity risk	Working with flashes, inversions, empty frames, strobing; seeing a one-frame item in critique
references/typography.md	Type weight, width, tracking, leading, combining multiple typefaces; enlargement, deformation, cropping, masking, repetition; type appearance and entrance	Selecting type, handling title text and lyrics; being asked about fonts
references/composition.md	Center of gravity, white space, symmetry vs. offset, hierarchy, alignment vs. offset; jobs of lines, dots, circles, stars, arrows; three grid types; eye movement; single frame vs. playback	Arranging a single frame's composition; critiquing layout; deciding whether to use geometric elements
references/stills-and-motion.md	What five images are in their final videos; the five things a still cannot tell; dense full-frame compositions hold only a few frames; designing for both playback and pause; matching an image to a video frame	When only stills are available; creating preview images or cover frames; making information-dense full-frame compositions
references/layout-and-type.md	Text to read vs. text to see, size and time, type size, small type, titles, lyric position, type tool survey, windows, frame, density, UI elements	Laying out frames; making covers or posters; multi-frame adaptation
references/color-and-texture.md	Color budget, four color uses, brightness difference, neutral colors, inversion, gradients & grain, texture families, transparent checkerboard, measuring color	Setting colors and texture; when colors mismatch; encountering symbols like checkerboard
references/space-2d-3d.md	Two relationships between 2D and 3D, annotation juxtaposition, cheap space, physical objects and re-photography	When the frame has 3D, live-action, or hand-drawn elements to be juxtaposed
references/structure-and-motif.md	Running motifs, axes, bookending, multi-use layout, production process as part of the shot, cover vs. final video, "job" correspondence table	Planning the entire piece; when the user wants to make something "like [that]"
references/cases.md	ID, specs, and reliable points of evidence for each work; relationship between 14 images and videos; constraints on citation	Before citing any case
references/analysis-method.md	Deconstruction sequence, scripts, metric definitions, errors made, note format, evidence levels	Deconstructing reference clips; measuring drafts; reading any numerical value
references/replication.md	Replication sequence and discipline, time and color pitfalls, metrics for similarity, record of rollback cases	Replicating a segment from reference

motion-design-judgment