cinematic-shot-direction

name	cinematic-shot-direction
description	Direct cinematic shots and coverage for AI-assisted content production. Use when an agent must translate story intent into framing, shot size, lens and perspective, camera position, movement, blocking, screen direction, lighting-aware coverage, continuity, shot specifications, or diagnose why generated or filmed shots do not cut together.

Direct cinematic shots as editorial decisions

Treat a shot as a controlled change in audience knowledge, attention, and feeling—not as a bundle of cinematic adjectives. Decide what the audience must perceive, choose the camera relationship that makes that perception likely, and specify only the variables needed to preserve it through production and editing.

Use three evidence labels throughout the work:

• Established fact: optics, geometry, or supported perceptual evidence.
• Production convention: a broadly used continuity or craft practice that may be broken deliberately.
• Production heuristic: a defensible starting point, not a law or guaranteed emotional effect.

Do not claim that a high angle always means weakness, a low angle always means power, a wide lens always feels energetic, or handheld always feels real. These are contextual readings. State the intended effect and test whether the complete image supports it.

Begin with the dramatic job

Write one sentence before choosing gear or movement:

At this beat, the audience should notice [subject/change], understand [relationship/geography], and feel [quality], while [information] remains concealed or secondary.

Then identify the shot's editorial function:

• Orient: establish people, objects, exits, distances, or direction of travel.
• Reveal: introduce new information by cutting, reframing, focus, blocking, or camera travel.
• Witness: hold a behavior long enough for performance and consequence to register.
• Identify: align with a character's looking, hearing, or bodily position.
• Compare: place subjects in one frame or matched frames so their relationship is legible.
• Isolate: remove contextual information to privilege face, gesture, texture, or object.
• Transition: carry direction, shape, action, occlusion, or attention into another shot.

If the sentence still works with a locked camera, prefer the locked camera until movement has a specific job. A moving camera can add meaning, but the ASC's professional guidance begins movement choice with the script and intended shot, not the rig (ASC, “Tools for Camera Movement”).

Build the frame from relationships

Choose shot size by information, not prestige

Shot-size boundaries are flexible; they describe how much of the subject and environment is visible, not an absolute camera distance (Oklahoma State University, “Cinematography”). Use the smallest scale that still contains every required piece of information.

Scale	Primary information	Frequent tradeoff
Extreme wide / establishing	geography, scale, destination, isolation	face and fine action become unreadable
Wide / full	full-body action, blocking, spatial relationship	subtle expression loses weight
Medium wide	body language plus meaningful environment	neither face nor geography dominates
Medium	gesture, interaction, conversational relation	feet, entrances, and larger geography disappear
Medium close-up	face with some posture and context	hand action and spatial relation may be lost
Close-up	expression, decision, tactile detail	geography and partner response disappear
Extreme close-up / insert	one decisive feature, object, or action	context must come from adjacent shots

Name the framing landmark when precision matters: “full figure with feet and doorway,” “mid-torso,” “shoulders and head,” or “eyes and mouth,” rather than relying on abbreviations alone. For objects, name the visible surfaces and scale cue.

Compose for a priority order:

1. Primary subject or action.
2. Secondary relationship needed for meaning.
3. Background evidence needed for place or consequence.
4. Negative space reserved for gaze, motion, graphics, or an upcoming reveal.

Inspect edges, occlusions, mergers, horizons, and bright distractions. Aspect ratio is part of the shot contract: Kodak recommends explicitly communicating the intended frame because downstream framing otherwise becomes guesswork (Kodak, “Aspect Ratio Targets”). For vertical and horizontal deliveries, redesign blocking and negative space; do not assume a center crop preserves the shot.

Separate distance, height, and angle

These are independent controls:

• Distance determines the camera's spatial relation to the subject and is the primary control of perspective.
• Height is the lens center's elevation relative to the subject: floor, waist, chest, eye, overhead.
• Angle is where the optical axis points or where the camera sits around the subject: level, looking up/down, frontal, three-quarter, profile, rear three-quarter.
• Roll is horizon tilt. Keep level unless imbalance or a graphic diagonal is intentional.

Specify both height and angle. “Low angle” is ambiguous: a camera near the floor can remain level, or an eye-height camera can tilt upward. The first changes foreground and horizon placement; the second changes convergence and which surfaces are seen.

Treat camera position as point of view. Ask who is permitted to see from here, what the position reveals, and whether the audience is observing, participating, pursuing, trapped, or judging. Do not reduce this to a fixed symbolism dictionary.

Use lens and perspective correctly

Established facts: focal length and sensor/image format jointly determine angle of view; focal length also changes magnification. Perspective—the relative apparent size and spacing of objects—is governed by camera position. “Telephoto compression” emerges when a longer lens is paired with a more distant camera position to maintain subject size, not from focal length alone (Sony, “Focal Length, Angle of View & Perspective”; Nikon, “Understanding Focal Length”).

Choose in this order:

1. Place the camera for the desired perspective and relational distance.
2. Choose the angle of view that includes the needed frame.
3. Choose aperture/focus behavior for attention, exposure, and practical focus tolerance.
4. Evaluate optical character only after the geometry works.

State focal length as a format-aware field of view: “moderate wide, about 28 mm full-frame equivalent,” not merely “28 mm.” If the capture format is unknown, specify “wide / normal / long field of view” plus desired distance and framing.

Practical relationships:

• Move close with a wide field of view to enlarge near/far differences, show more environment, and make camera presence felt.
• Move back with a long field of view to flatten relative depth, simplify background coverage, and observe from a distance.
• Hold camera position and change focal length to crop or widen without changing perspective.
• Keep subject size but change focal length and distance together to change perspective.
• Treat shallow focus as an attention tool with continuity cost. Depth of field changes with camera distance, focal length, and aperture (Sony, “What is depth of field?”).

The ASC advises choosing lens, camera placement, and key light consciously as a system, and notes that a wide lens can support multiple actions in one frame while a longer lens can isolate a subject from farther away (ASC, “Fresh Perspectives”).

Choreograph movement and blocking together

First choose what moves:

• Subject only: the frame becomes a stage; blocking creates emphasis and depth.
• Camera only: the viewer's relationship changes while action holds.
• Subject and camera together: tracking can preserve size while backgrounds and relationships evolve.
• Neither: duration, performance, light, and sound carry the beat.

Name camera movement geometrically and dramatically:

Move	Geometry	Useful dramatic job
Pan / tilt	rotate from a fixed position	redirect attention, connect offscreen space, follow action
Dolly in/out	translate toward/away	cross emotional or informational distance; reveal parallax
Truck/track	translate sideways	follow travel, expose layers, compare subjects
Pedestal/boom/crane	translate vertically	reveal hierarchy or geography through changing height
Arc/orbit	travel around subject	change background, alignment, or relational side
Handheld	operator-borne micro-translation/rotation	embody instability, urgency, proximity, or human observation
Stabilized follow	smooth free translation	accompany a subject through space without dolly rigidity
Zoom	change focal length from fixed camera position	change framing/angle of view without translation or parallax
Rack focus	change focus plane, not camera position	transfer attention between depth layers

Specify a movement with start frame, path, motivation/cue, speed character, end frame, and what becomes newly visible. “Slow cinematic push-in” is incomplete. Prefer: “Start medium two-shot; when Mina hides the key, advance 60 cm on axis over four seconds, ending chest-up on Mina while Theo falls soft at frame edge; no pan.”

Block subjects with marks and beats:

• starting and ending positions;
• facing and eyeline target;
• path, pace, and action cue;
• foreground/background crossings;
• moments of occlusion or reveal;
• required separation and silhouette;
• relation to focus plane and light.

Rehearse camera, subjects, focus, and light as one system. A subject stepping toward the key may brighten and become frontal; crossing behind it may lose eye light; an arc can turn modeled side light into flat frontal light or silhouette. ARRI distinguishes key as the primary modeling source and fill as shadow control, while warning that hard opposing fill can create unnatural double shadows (ARRI Lighting Handbook). Track key side, softness, contrast, color, background level, practical positions, and motivated changes across coverage.

Preserve screen direction and editorial continuity

Define the active axis before coverage:

• two-person dialogue: line between the people;
• travel: direction of motion;
• look: line from observer to object;
• task: line connecting person, hands, and work surface.

Production convention: keeping cameras on one side of the axis preserves left/right placement, opposing eyelines, and direction of travel. Crossing may reverse those relationships (Columbia Film Language Glossary, “180-Degree Rule”; BBC, “Crossing the Line: Movement”). The convention is not a moral rule. Cross deliberately when reversal, rupture, or reorientation is the point, and bridge it with visible camera travel across the axis, a neutral/on-axis view, or a new establishing relation.

For each shot, record:

• who is screen-left and screen-right;
• gaze direction and exact eyeline target;
• entry and exit edge;
• motion direction;
• axis side and whether/where it changes;
• hand, prop, wardrobe, and action state at the cut;
• focus target and lighting state;
• start and end pose with handles.

Use match on action to carry attention and causal motion through a cut. Peer-reviewed evidence supports the idea that continuity techniques guide attention and reduce awareness of cuts; action coinciding with a cut is particularly effective (Smith & Henderson, 2008; Smith, 2012). An eyeline match links a look to its object (Columbia Film Language Glossary, “Eye-Line Match”). Preserve the causal order: look → object, action → consequence, question → reaction.

Design coverage around beats, not insurance

Coverage exists to construct the scene's intended rhythm and emphasis. Do not collect every subject at every size by default.

For each dramatic beat, determine:

1. Whose information or reaction controls the beat?
2. Must audience geography remain visible?
3. Is the decisive event simultaneous with another event?
4. Does performance need uninterrupted duration?
5. Where can an editor enter and leave the shot?

A robust minimal set often includes:

• one orienting or relationship view;
• one view for each decisive action or speaker;
• one reaction that changes interpretation;
• inserts only for story-critical details;
• an exit, transition, or re-establishing option;
• clean action overlap and handles.

The BBC's five-shot guidance—face close-up, action close-up, two-shot, over-shoulder, and an unusual “art” angle—is useful as a coverage check, not a mandatory formula; it also advises a meaningful change in both size and angle between adjacent shots (BBC, “Five Essential Shots”).

If a beat depends on simultaneity, preserve it in a two-shot, deep composition, split attention, or choreographed oner rather than fragmenting it into singles. If a performance depends on escalation, allow a sustained shot and reserve coverage for the actual turn. Coverage can live inside a moving master when blocking, focus, and exposure support it.

Write a generative production shot specification

Describe one continuous shot per specification unless the production system explicitly supports independently controllable multi-shot output. Keep story context outside the visual instructions when it does not affect what appears.

Use this order:

SHOT ID / EDITORIAL FUNCTION
Intent: what changes for the audience in this shot.
Duration / delivery: approximate usable duration, frame rate if important, aspect ratio and crop obligations.
Subject and setting: visible identities, objects, surfaces, time, weather, and spatial anchors.
Start frame: shot size, composition, subject placement, camera distance, height, angle, roll.
Lens / focus: format-aware field of view, focus target, depth behavior, intentional breathing or distortion only if relevant.
Blocking: start mark, action path, eyeline, gesture/prop state, end mark.
Camera movement: geometry, path, cue, speed profile, stabilization character, end frame.
Lighting interaction: motivated source direction, key side, contrast, practicals, atmosphere, and how movement changes light.
Continuity: axis side, screen direction, entry/exit, previous and next action states, match point.
Exclusions / tolerances: forbidden extra subjects, text, cuts, camera operations, or changes; acceptable variation.
Review target: the two or three conditions that make the take usable.

Prioritize a single dominant subject action and a single dominant camera idea. This is a production heuristic for controllability, not a provider limit. If two moves are essential, state their sequence and cue rather than listing them as simultaneous style words.

Diagnose failures from evidence

Review the actual frames or clip before rewriting direction. Identify the first frame where intention breaks.

Symptom	Likely production cause	Repair
“Long-lens compression” is absent	focal length changed but camera position did not	move camera back, restore subject size with longer field of view, preserve background anchors
Face is distorted or room stretches	camera too close for intended perspective	move back; select narrower field of view; restage foreground
Shot feels unmotivated	movement lacks a story cue or reveal	lock camera, or tie move to a precise action/information change
Subject drifts or scale pumps	camera and blocking lack start/end geometry	specify marks, path, screen position, end frame, and whether size is held
Arc becomes an accidental reverse	camera crosses axis without reorientation	stop before axis, show the crossing, or establish the new side
Dialogue partners appear to look the same way	eyelines or screen sides are unspecified/inconsistent	define axis side, screen placement, and offscreen targets for both singles
Coverage will not cut	adjacent frames are too similar, action states mismatch, or no handles	change size and angle meaningfully; overlap action; record start/end states and handles
Lighting jumps between angles	key side, practical state, or time-of-day logic changed	map motivated sources; preserve key direction/contrast; flag intentional changes
Face loses modeling during track	subject/camera path moved relative to key	adjust blocking, carry/broaden source, or embrace and cue the transition
Background steals attention	bright/color/contrast/motion hierarchy conflicts with intent	reframe, simplify, dim, defocus, or time background movement
Generated shot invents a cut	specification contains multiple shots or sequential viewpoints	split into one take per spec; state “single continuous shot, no cuts”
Detail is unreadable	shot size, duration, focus, or occlusion does not support it	allocate an insert or hold longer with a clean focus/lighting path

Change one causal variable at a time when practical. Preserve what already works: identity, blocking, screen direction, palette, or performance. Do not solve a geometry failure with more mood adjectives.

Example 1 — dialogue reveal with motivated coverage

Example status: illustrative production example, not a mandatory formula.

Intent and constraints: In a quiet kitchen, Mina hides a stolen key while Theo asks an innocent question. The audience must see the concealment before Theo does. Keep the scene readable, restrained, and editable in 16:9.

Axis and lighting plan: Establish the axis between Mina at the counter on frame left and Theo in the doorway on frame right. Keep dialogue cameras on the room side of that line. Window camera-left motivates a soft side key; warm pendant remains visible but does not reverse the key. Mina's hand can enter a darker area under the counter, making concealment plausible without making the key unreadable.

Coverage plan:

1. 1A relationship master — medium-wide two-shot, both faces and counter visible; hold Theo's entrance, question, and Mina's first reaction.
2. 1B action insert — close insert of Mina's hand sliding the key beneath a folded towel; match the hand action from 1A.
3. 1C Mina single — medium close-up, Theo's soft shoulder at frame right; push only after the key disappears.
4. 1D Theo reaction — medium close-up, looking frame left; remain static so his lack of knowledge reads as calm rather than suspicion.
5. 1E re-establish — return to two-shot as Mina turns away, providing a scene exit.

Complete shot specification for 1C:

SHOT 1C / REVEAL MINA'S DECISION
Intent: transfer attention from the hidden key to Mina's controlled lie; audience should feel proximity but not accusation.
Duration / delivery: 6 seconds usable, 16:9, protect a centered 4:5 crop without losing Theo's shoulder.
Subject and setting: Mina at a dark wood kitchen counter at blue hour; Theo stands in the doorway behind camera-right; folded towel and warm pendant remain continuity anchors.
Start frame: medium close-up from mid-torso, Mina left-of-center with look room right; camera 15 degrees off frontal on the room side of the dialogue axis, lens at upper-chest height and level.
Lens / focus: normal-to-mild-long field of view, approximately 65 mm full-frame equivalent; focus on Mina's near eye; Theo's shoulder is soft but recognizable.
Blocking: start as Mina's right hand completes the concealment below frame; she raises her eyes to Theo without turning her torso, pauses, then says, “What key?”
Camera movement: on the upward eye movement, dolly forward about 45 cm over 3 seconds; no pan, tilt, orbit, or zoom; finish shoulders-and-head with stable look room.
Lighting interaction: soft window key remains camera-left and slightly behind Mina; maintain eye light through the dolly; pendant warms the background only; do not flatten her face at end mark.
Continuity: Mina remains screen-left and looks frame-right; concealment is complete before shot starts; cut in on her hand leaving frame and cut out after her line and one-second hold.
Exclusions / tolerances: single continuous shot, no extra people, no camera shake, no rack focus, no change in weather or practical state. Natural micro-expression is welcome.
Review target: correct eyeline, eye-motivated push, and uninterrupted light modeling across the move.

Likely failures and variation: If the push starts before her eyes rise, it announces importance too early; delay the cue. If the scene should feel observational, move the camera farther back with a longer field of view and remain locked. If Theo is meant to suspect her, hold him sharp at frame edge or give him the final reaction instead.

Example 2 — product reveal driven by parallax and light

Example status: illustrative production example, not a mandatory formula.

Intent and constraints: Reveal a brushed-metal watch as engineered rather than luxurious. One eight-second 9:16 shot; no hands, text, or impossible floating parts.

SHOT P1 / REVEAL CONSTRUCTION
Intent: make the audience discover the crown, layered case, and face in that order through camera parallax and a controlled highlight.
Duration / delivery: 8 seconds, vertical 9:16; all critical product edges inside the central 80 percent.
Subject and setting: one brushed titanium watch upright on a matte charcoal plinth; fine machining remains physically plausible; dark neutral studio with faint haze only in background.
Start frame: macro close-up on the crown in right profile, crown at lower-right third, dial mostly hidden by the case edge; camera slightly below dial center and level.
Lens / focus: narrow macro field of view with natural perspective from a moderate working distance; focus begins on crown teeth, then transfers once to the near dial markers.
Blocking: watch and plinth remain completely stationary.
Camera movement: smooth 35-degree leftward arc around the watch over 7 seconds, constant radius and height, easing gently at start and stop; reveal full dial by second 5; no zoom or roll.
Lighting interaction: large cool soft source from rear-left creates a broad traveling edge highlight as the camera arcs; narrow warmer card reflection defines crown; black negative fill camera-right preserves metal shape. Highlight must travel continuously, never clip to featureless white.
Continuity: single self-contained shot; end frame holds one second for a possible title overlay above the watch.
Exclusions / tolerances: no rotation of watch, no melting geometry, no duplicate crowns, no extra jewelry, no engraved brand text, no cuts. Minor dust-free reflection variation acceptable.
Review target: mechanically stable silhouette, crown-to-dial reveal order, and continuous controlled highlight.

Why structured this way: The arc changes visible surfaces and creates parallax; a zoom would only change framing. Locking the object isolates camera-motion errors. The lighting direction is specified relative to object and path so the reveal does not destroy form.

Likely failures and variation: If the face seems to rotate independently, reinforce “stationary watch” and describe fixed background anchors. If metal reads plastic, broaden the source reflection and deepen negative fill rather than adding “premium.” For a precision-documentary version, remove haze and rack focus, use deeper focus, and end square to the dial.

Example 3 — travel continuity across generated shots

Example status: illustrative production example, not a mandatory formula.

Intent and constraints: A courier crosses a flooded street toward a red pharmacy sign. Three separately generated shots must cut as one journey; the audience must never think the courier turned back.

Sequence contract: Travel direction is screen-left to screen-right in all three shots. Rain falls with the same wind direction. Courier wears yellow shell, black backpack, and carries the parcel under the left arm. The pharmacy sign stays the destination and grows larger across the sequence.

1. W1 orient: wide lateral view, courier enters left and exits right; pharmacy sign small at upper-right; locked chest-height camera; deep focus.
2. W2 accompany: medium-wide parallel track on the same side of motion, holding courier at left third while background moves right-to-left; moderate wide field of view; parcel remains under left arm.
3. W3 arrive: low medium frontal three-quarter from destination side but still respecting the established axis; courier moves diagonally left-to-right and stops beneath the now-large sign; camera eases backward without crossing the path.

Match points: Cut W1→W2 as the right foot strikes water; both shots show the same right-foot contact and splash direction. Cut W2→W3 as the courier lifts eyes toward the sign; W3 begins with that upward gaze and preserves the sign's screen-right relation.

Critical failure check: A right-to-left exit or opposite wind in any shot implies reversal. A true head-on neutral shot could bridge an axis change, but an unannounced reverse angle cannot.

Final review

Before approving a shot or sequence, verify:

• The dramatic job is legible without the shot description.
• Required information is visible at the chosen size and duration.
• Camera position creates the intended perspective; focal length is format-aware.
• Height, angle, and roll are explicit rather than conflated.
• Movement has a cue, path, end frame, and reveal or relational purpose.
• Blocking, focus, and lighting remain compatible through the full take.
• Axis, eyelines, screen direction, action state, and props connect to adjacent shots.
• Coverage expresses the beats and supplies actual edit points.
• Aspect-ratio variants preserve subject, gaze/motion room, and important context.
• The specification contains one controllable camera idea and states tolerances.
• Review notes distinguish established fact, convention, and heuristic.

Handoff into observable video language

Shot direction says what production should create; downstream logging, training, retrieval, and prompt comparison need a separate statement of what the finished shot should make observable. After approving a shot, add a compact five-lane handoff:

• Subject: stable visible entities, distinguishing attributes, pose, relationships, and intended start/end state.
• Scene: setting, overlays, time/weather evidence, and visible context that must survive production.
• Motion: actor, action, target, direction, contact, and temporal order.
• Spatial: shot size, frame position, depth, occlusion, screen direction, and start/end framing.
• Camera: height, angle, field-of-view character, focus, steadiness, movement geometry, and transition boundaries.

This handoff is not a generated description of the final footage and must not be marked as observed fact. Once footage exists, use precise-video-description to describe the actual result and preserve discrepancies from the intended spec. Use video-description-oversight when source descriptions, shot logs, or generation metadata need evidence-backed correction and acceptance.

Keep terminology consistent across a sequence. If “dolly,” “zoom,” “frame-left,” “subject-left,” “medium shot,” or another term is used differently between shot specifications, resolve the glossary before provider prompting or dataset handoff. Lin et al.'s CHAI research reports that structured Subject, Scene, Motion, Spatial, and Camera coverage improves consistency and exposes omissions, but its dataset-specific caption lengths and primitive inventory are not universal production requirements.

Sources

Sources accessed 2026-07-09. These are stable craft, optics, and research references; no provider capabilities or volatile model claims are used.

• American Society of Cinematographers: “Fresh Perspectives” and “Tools for Camera Movement”.
• ARRI: Lighting Handbook.
• BBC Academy: “Five Essential Shots” and “Crossing the Line: Movement”.
• Columbia University Film Language Glossary: “180-Degree Rule” and “Eye-Line Match”.
• Eastman Kodak: The Essential Reference Guide for Filmmakers and “Aspect Ratio Targets”.
• Oklahoma State University Pressbooks: “Cinematography” and “Editing”.
• Smith, T. J. (2012): “The Attentional Theory of Cinematic Continuity”, Projections, 6(1).
• Smith, T. J., & Henderson, J. M. (2008): “Edit Blindness”, Journal of Eye Movement Research, 2(2).
• Sony: “Focal Length, Angle of View & Perspective” and “What is depth of field?”.
• Nikon: “Understanding Focal Length”.
• Lin et al., “Building a Precise Video Language with Human-AI Oversight”, CVPR 2026, checked 2026-07-14.