Chapter 29. From Beat to Shot#

29.1 A beat defines change; a shot decides how it is watched#

The same beat can be carried by one long take, three exchanges of shot and reverse, or a set of proof inserts. The job of shot design is not to illustrate the script sentence by sentence. It is choosing what the audience looks at first, when they understand it, and which expression they stay on.

Every shot must answer what would be lost if it were deleted. The answer might be information, action, emotion, space, proof, rhythm, or an editing interface.

29.2 Six shot functions#

Establishing shots state space and relationships. Action shots show a behavior's start and end. Information shots let the audience read a new fact. Emotion shots show a character's judgment. Consequence shots land the change in power. Transition shots connect time, place or action.

One shot can serve several, and the shot list must record the primary function. A beautiful shot with no function only adds generation, continuity and editing cost.

29.3 Derive coverage from the state change#

The state change in E001_B04 is that the authorization gives Lin Yun standing. At minimum you need: Lin Yun putting the document into public space; the authority being legible; a key person confirming it; Lin Wei's reaction; the meeting's behavior changing.

coverage:
  beat: B04
  shots:
    - E001_S08: medium, Lin Yun pushes the authorization to the center of the table
    - E001_S09: document insert, the title and signature area
    - E001_S10: the chair's hand and eyes confirming
    - E001_S11: Lin Wei close-up, the smile stops
    - E001_S12: short wide, security steps back and Lin Yun moves to the table

Proof, reaction and consequence all have a place, so the reversal is not merely a line.

29.4 Using 12–15 shots per minute correctly#

That range is common production experience, usable for a first budget. It is not an aesthetic law. Roughly 15–19 shots in 75 seconds suits dialogue- and proof-driven microdrama. Fast explainer formats can run more; intimate performance can run fewer.

Shot count is set by information units, action risk, performance pauses and how long the material stays stable. A model that can only hold three seconds of action may require more shots — but you must not chop up an emotion to suit a tool.

29.5 Shot economy#

A shot that only repeats known information and provides neither a reaction nor cover for the edit should be deleted. Conversely, proof objects and character reactions cannot both be omitted in the name of pace.

Grade expensive shots. Hero shots carry the hook, the reversal, the cut point or the core performance. Functional shots secure the causality. Coverage shots give the edit choices. Insurance shots guard against failure. Budget protects hero and functional shots first.

29.6 Design the master first, then the coverage#

The master establishes the geography and the complete action; coverage emphasizes performance, evidence and reaction. AI production need not literally generate a long master first, and shot design must still know how the whole event occupies the space.

Otherwise several attractive close-ups cannot assemble into one action, and the edit is reduced to explaining it in narration.

29.7 The three-shot structure for proof#

Proof-driven microdrama commonly uses: the character showing intent; a clean insert of the object; the opponent's or the public's reaction. Add the consequence of its use where needed. Shooting only the document loses the emotion; shooting only the reaction loses the credible fact.

Composite important text separately, and keep the insert frontal, stable and long enough. The shot of a character holding it carries the relationship; it does not have to carry the full read.

29.8 Choosing the reaction shot#

Reaction shots do not distribute attention evenly. Choose whoever loses most, knows the secret, or changes sides. After the authority is confirmed in E001, Lin Wei's smile stopping is worth far more than an ordinary director looking surprised; security stepping back is the institutional consequence.

Reactions can also cover discontinuity between action and background, and they need a narrative reason first.

29.9 Nine-panel and twenty-five-panel grids#

A nine-panel grid quickly checks composition, shot sizes and space across one section. A twenty-five-panel grid checks an episode's rhythm and continuity. Grids are review tools, not a shortcut that lets a model invent every precise shot at once.

Every cell must map to a shot ID. If characters and locations in a grid have not passed asset review, they cannot be used directly as video input. Fix the individual cell first, then move to keyframes.

29.10 How the shot list relates to the grid#

The shot list holds precise state, assets, sound and intent. The grid gives a visual overview. Changing a composition in the grid must be written back to the shot list, or you have two truths.

The shot list is authoritative data; the grid is a rendered view. Grid titles and numbering can be generated from the shot list automatically.

29.11 Auditing shot variety#

Across fifteen consecutive shots, avoid meaninglessly repeating one size, angle or camera move more than about three times — and every change needs a narrative reason. Rising power can move from a compressed close-up to a steadier medium; a reveal can move from person to proof to consequence.

Changing angles at random breaks the axis and the style. Variation is not entertainment; it is control of attention.

29.12 The shot row#

shot_id: E001_S09
beat_id: B04
function: proof_insert
duration: 2.6s
shot_size: extreme_close_up
angle: top_down_10deg
camera_move: slow_micro_push
subject: PROP_AUTHORIZATION
action: the document is pushed in by the right hand and comes to rest
state_in: the edge of the document enters frame
state_out: title and signature area stable and readable
continuity_refs:
  assets: [PROP_AUTHORIZATION@v03, GRAPHIC_AUTHORIZATION_V03]
audio_intent: paper friction, ambience drops
edit_intent: cut in after the stress in Lin Yun's line, out to the chair confirming

29.13 SOP from beat to shot#

First, confirm the beat's entry and exit states. Second, list the information, action and consequence the audience must see. Third, choose the master. Fourth, add proof, performance and reaction coverage. Fifth, write a primary function for each shot. Sixth, estimate duration and risk. Seventh, check the spatial, action and audio interfaces. Eighth, produce a nine- or twenty-five-panel overview. Ninth, delete duplicate shots and protect the necessary reading time.

29.14 Fault tree#

Symptom: one image per line of dialogue. Shots were cut from text rather than state change. Return to beats and shot functions.

Symptom: many shots and the reversal is unsatisfying. Proof, reaction or public consequence is missing. Complete the three-shot structure.

Symptom: the grid is beautiful and cannot generate continuous video. The cells have no stable IDs or state. Fix individual cells and write back to the shot ledger.

Symptom: everything is a medium close-up. There is no hierarchy of information, space and consequence shots. Adjust sizes by function.

Symptom: the axis is crossed constantly for variety. Random angles have replaced narrative change. Lock the axis and re-establish only when power or position changes.

29.15 Checklist, exercises and deliverables#

Check that each shot has a primary function; that beats cover action, proof and consequence; that shot count follows content; that the master establishes geography; that reactions choose the right person; that grid cells reference shot IDs; that duplicates are deleted; and that hero shots receive the budget.

Exercise one: design one beat as a one-shot, a three-shot and a five-shot version. Exercise two: build a nine-panel grid for the authorization reversal. Exercise three: delete half the shots and state which information cannot be lost. Exercise four: count the distribution of shot functions across one minute.

Deliverables for this chapter: the coverage plan, the shot ledger, shot cost tiers, the nine-panel grid, the twenty-five-panel grid, the shot variety audit, and the deleted-shot record.

29.16 The coverage compiler: from a change list to shot requirements#

Each beat first lists five kinds of change: fact, power, emotion, space and physical state. The coverage compiler does not decide shot sizes directly; it generates the requirement of what the audience must see. The authorization being confirmed changes the fact, the power at the door, and security's behavior at once — so at minimum you need a legible document, the confirmer's reaction, and the path opening. One shot may carry several outcomes, and every outcome must have a shot responsible for it.

Figure 29-1 Changes in fact, power, emotion, space and physical state become "must be observed" requirements, which the shot pack then covers item by item. A requirement no shot is responsible for is a gap; a shot carrying the same information with no new consequence is redundant.

The ticks in the diagram do not demand one shot per class of change. A document close-up can prove both the fact and the prop state; a master can show space and power together. Shot economy comes from merging responsibilities — and after merging, everything must still read clearly at phone size.

coverage_requirement:
  beat: B04_authorization_confirmed
  changes:
    fact: the authorization is valid
    power: Lin Wei loses control of the doorway
    spatial: the path changes from blocked to open
  must_observe:
    - the signature or confirmation action
    - Zhou Lan making an institutional judgment
    - security standing down
  optional:
    - Lin Wei's smile stopping
  forbidden_substitute:
    - Lin Yun simply announcing the result in dialogue

Once shots are designed, check them back against the requirements. A requirement with no owner is a gap. Several shots serving one requirement with no new reaction or consequence is redundancy. "Do we have enough shots" then stops being estimated from experience and is proven from the story's changes.

29.17 Shot IDs, dependencies and replacement boundaries#

Every shot in the shot ledger holds a stable ID. Version changes do not change the ID; only a fundamental change of narrative function creates a new one. Record what it reads — beat, state, characters, location, props, dialogue and sound events — and what it outputs to the next shot in action, eyeline and emotional state.

Before replacing a shot, check the boundary contract: does the new shot still supply the same fact, does it enter and leave in compatible states, does it retain the necessary handles. A more attractive frame with a different output state is not v2 of the same shot. The system should mark downstream dependencies for regression.

Stable IDs also let editing, subtitles, music and QC share one reference. A comment reads E001_S014@v06 00:01.2 rather than "that document shot" — so context survives changes to filenames and timeline positions.

29.18 An animatic is a production test, not a presentation#

Once the grid passes, build an animatic from stills, scratch dialogue, rough effects and estimated durations. The purpose is not to simulate finished quality. It is to test information order, eyelines, shot count, evidence reading time and cut point position. If a cheap animatic is already incomprehensible, expensive generation will not fix it.

Run the animatic at least three times: muted for space and causality; eyes closed for dialogue and sound events; at phone size for characters, text and subtitles. The director records decisions to delete, merge and add shots, and updates the shot ledger — rather than treating the animatic as a one-off video.

Before generation, run a shot preflight: every required asset approved, no conflicts in the state inputs, complete contracts with the shots either side, alternatives available for risky shots, and estimated generation plus repair cost within limits. A shot that fails preflight does not enter the queue, which stops the model from making upstream decisions on the team's behalf.

A note on sources#

Chinese AI shot-design practice commonly uses nine-panel and twenty-five-panel grids and roughly 12–15 shots per minute. This chapter positions those as planning and review tools rather than a one-click substitute for shot state and asset management.