skills/film-scenes/references/PRESETS-CAMERA.md

# Camera & optics presets — movement, size, FOV, aperture, angle, view, arrangement, placement, composition

> **The wordings moved to the server (2026-09-14).** This file still decides
> *which* preset to reach for — that judgement is the part that needs a reader.
> What it no longer carries is the sentence each id compiles to: those were
> tuned against real renders, and a prompt written from a remembered copy of
> one renders worse than a prompt that pastes it. Pass the id to
> `picsart_film_compile_prompt` and the server pastes the sentence.
>
> This is not a style rule. Measured on a real shipped run, every one of seven
> shots had its size, angle and field of view reworded in exactly this way.


The vocabulary behind the shot-level camera decisions and their text-choice
fallback. Movement, size and angle open every shot block (`PROMPT-BLOCKS.md`,
Part 2 — *shot type, camera angle, camera movement*); FOV and aperture belong
to each shot and reach its prompt on **every** shot, whether the user changed
them or left them as seeded (*The card's own picks*, revised 2026-09-30);
arrangement, character view, placement and
composition compile into the scene picture's prompt and the geography line.
(The block numbers this file used to cite belonged to the fifteen-block
template, gone 2026-09-09.) Fragments are pasted verbatim;
`<subject>`, `<second>`, `<third>` are replaced with asset tags' names. Since
2026-09-08 the tables carry the Zero Shoots handbook's *Camera Work* vocabulary
name for name — its phrases are the fragments wherever it has one.

The **movement** table below is served by every console; the rest of this file
is authored from here by hand, because the retired `picsart_shot_designer`
console was the only surface that ever drew it
(`../../../docs/widget-specs/shot-design-panel.retired.md`).

## Movement presets — one per shot

**50 moves in six groups, and they are one library.** The same ids, labels and
notes serve the shot list's `Camera move` control, the review board's direction
panel, the film-setup console's world default, and the retired shot designer's
Movement tab if it is ever re-registered. A move means the same thing wherever
the film is discussed, which is the whole reason a note reading
`Camera move → Slow push-in` can be folded into a prompt without asking what the user
meant by it — and why picking one without boards, with no console at all, lands on
the same fragment the board would have compiled.

Grouped by energy. Every fragment also states what the camera does NOT do —
models drift without the negative space written positively. Three laws across
the whole table:

- **Every move ends somewhere named.** The fragment's last clause is the END
  STATE — the final frame the move arrives at ("ending composed on…",
  "settling at…"). Models overshoot moves with no destination, and the end
  state is what the next shot's seam inherits. The exception is a move that is
  *defined* by not arriving — a sustained track, a chase, an infinite zoom —
  where the last clause names the constant instead ("constant distance",
  "never leaving frame").
- **Speed is a variant, not a new move.** "Faster" or "slower" modifies the
  chosen preset's fragment in place (one word), never switches presets. Where
  the table does name a speed (slow / fast / crash zoom) those are three
  genuinely different shapes — a settle, an arrival, an overshoot — not three
  dials on one move.
- **Direction is a placeholder, not a preset.** `<left/right>` and `<up/down>`
  are filled at prompt time; a pan left and a pan right are one move pointed
  two ways. Surfaces that cannot ask (the shot list, the world default) drop
  the word rather than guess it.

### At rest (register: SETTLE)

**Ids:** `locked off` · `handheld settle` · `slow push-in` · `slow pull-back` · `slow zoom in` · `slow zoom out` · `slow pan` · `slow tilt` · `pedestal` · `slider`

*(The sentence each id compiles to now lives server-side. Pass the id to `picsart_film_compile_prompt` and it pastes the tuned wording; never write that sentence yourself.)*

**Why there is no `arc`, either.** It was a quarter-circle around a static
subject — which is what `orbit` is, and the probe confirmed a model cannot be
asked to tell them apart: generated as a deliberately shallow 30° swing it came
back as plain lateral travel, indistinguishable from a slider and nothing like a
curve around anybody. One circling move is enough. `arc` resolves to `orbit`.

**Slider against pan, and why there is no `truck`.** Both move the frame
sideways and a model will collapse them if the fragment lets it: a pan pivots
from one spot and has no parallax, while a slider is a short move that re-frames
the same subject.

A third preset — `truck`, real travel across the scene — was carried here
briefly and **retired by decision**. On paper it is distinct: lateral travel past
a *static* subject, where `tracking lateral` is the same path *alongside a moving
one*. In practice the two compile to the same camera and the difference lives
entirely in whether the subject happens to be walking, which the shot already
says. One preset for lateral travel is enough; `truck` resolves to
`tracking lateral`.

### In motion (register: kinetic)

**Ids:** `tracking lateral` · `low tracking` · `follow behind` · `front tracking` · `handheld run` · `handheld shaky` · `chase` · `orbit` · `roll` · `dolly counter` · `crane up` · `crane down` · `push through` · `vehicle track` · `pass-by reveal`

*(The sentence each id compiles to now lives server-side. Pass the id to `picsart_film_compile_prompt` and it pastes the tuned wording; never write that sentence yourself.)*

**The three tracking moves are named by WHERE THE CAMERA IS**, because that is
the only thing that separates them: `tracking lateral` rides beside the subject,
`front tracking` runs ahead of them facing back, and `follow behind` sits behind
their shoulder. All three match the subject's pace and hold them in the same
place in frame; a viewer tells them apart by which side of the person they are
looking at. `front tracking` was called `lead walk` until the names were made to
say that — it still resolves.

**Chase against handheld run.** `handheld run` is a body carrying the camera —
the operator's effort is the texture. `chase` is about the route and the gap:
it can be a car, a drone or a gimbal, and what it promises is that the subject
stays in frame while the distance keeps changing.

### Punctuation (use sparingly — one per scene at most)

**Ids:** `punch-in` · `whip pan` · `fast zoom in` · `fast zoom out` · `crash zoom` · `crash zoom out` · `snap tilt` · `drop reveal` · `through glass`

*(The sentence each id compiles to now lives server-side. Pass the id to `picsart_film_compile_prompt` and it pastes the tuned wording; never write that sentence yourself.)*

**The three zoom speeds are three shapes.** `slow zoom` settles, `fast zoom`
arrives and holds, `crash zoom` overshoots and recovers. That is why the speed
law does not apply to them: writing "faster" onto a slow zoom gets you a fast
zoom's timing with a settle's ending, which is neither.

### Point of view

**Ids:** `POV walk` · `POV vehicle` · `mirror/insert POV` · `snorricam`

*(The sentence each id compiles to now lives server-side. Pass the id to `picsart_film_compile_prompt` and it pastes the tuned wording; never write that sentence yourself.)*

### Aerial / remote

**Ids:** `drone establish` · `drone track` · `drone push-in` · `drone pull-back` · `helicopter pull` · `robot arm`

*(The sentence each id compiles to now lives server-side. Pass the id to `picsart_film_compile_prompt` and it pastes the tuned wording; never write that sentence yourself.)*

**Drone pull-back against helicopter pull.** The drone move holds its altitude
and its speed; the helicopter move accelerates and is about scale arriving. Ask
which one the shot is doing before picking, because they read differently at the
same distance.

### Specials — the ones that are not really a camera path

| Preset | Register |
| --- | --- |
| rack focus | SETTLE |
| infinite zoom | kinetic |
| earth zoom out | kinetic |
| time-lapse | SETTLE |
| tilt-shift glide | kinetic |
| pass through | kinetic |

*(The sentence each id compiles to now lives server-side. Pass the id to `picsart_film_compile_prompt` and it pastes the tuned wording; never write that sentence yourself.)*

Two of these carry their own **register** because the group's does not fit them:
`rack focus` and `time-lapse` are shot on a camera that does not move, and a
Camera block reading `Register: kinetic` over a locked-off frame is a
contradiction the model has to resolve on its own. Every other preset takes its
group's register.

`pass through` is the general case of `through glass`: glass is the one that
works most reliably, so it keeps its own preset and stays in punctuation, and
`pass through` is what you reach for when the barrier is a wall or a hedge.

### Coverage — the aicameramovements.com set

The public list at `aicameramovements.com` is 46 moves in seven categories, and
every one of them is in the table above. Most were already here under the name a
crew actually uses; **11 were missing and were added** (marked ✚). Directional
pairs are one preset with a `<left/right>` placeholder, which is why 46 site
entries map onto fewer rows than 46.

| Site category | Site name | Preset here |
|---|---|---|
| Pan/Tilt | Static shot | `locked off` |
| Pan/Tilt | Pan right · Pan left | `slow pan <left/right>` |
| Pan/Tilt | Whip pan right · Whip pan left | `whip pan <left/right>` |
| Pan/Tilt | Tilt up · Tilt down | `slow tilt <up/down>` |
| Zoom/Lens | Slow zoom in | `slow zoom in` |
| Zoom/Lens | Slow zoom out | `slow zoom out` |
| Zoom/Lens | Fast zoom in | ✚ `fast zoom in` |
| Zoom/Lens | Fast zoom out | ✚ `fast zoom out` |
| Zoom/Lens | Crash zoom in | `crash zoom` |
| Zoom/Lens | Crash zoom out | ✚ `crash zoom out` |
| Dolly/Track | Dolly in | `slow push-in` |
| Dolly/Track | Dolly out | `slow pull-back` |
| Dolly/Track | Tracking shot | `front tracking` |
| Dolly/Track | Side tracking | `tracking lateral` |
| Dolly/Track | Follow shot / over-the-shoulder | `follow behind` |
| Dolly/Track | Reverse tracking / walk-and-talk | `front tracking` |
| Dolly/Track | Low tracking | ✚ `low tracking` |
| Dolly/Track | Vehicle tracking | `vehicle track` |
| Dolly/Track | Chase shot | ✚ `chase` |
| Physical | Truck right · Truck left | `tracking lateral` |
| Physical | Pedestal up · Pedestal down | `pedestal` |
| Physical | Slider right · Slider left | `slider <left/right>` |
| Physical | Push past / pass-by shot | `pass-by reveal` |
| Physical | Arc right · Arc left | `orbit` |
| Physical | Orbit clockwise · counterclockwise | `orbit` |
| Human | Handheld shot | `handheld settle` (running: `handheld run`) |
| Human | Body-mounted camera / Snorricam | `snorricam` |
| Drone/Crane | Crane up | `crane up` |
| Drone/Crane | Crane down | `crane down` |
| Drone/Crane | Drone push in | ✚ `drone push-in` |
| Drone/Crane | Drone pull back | ✚ `drone pull-back` |
| Drone/Crane | Helicopter shot | `drone establish` (accelerating: `helicopter pull`) |
| Specials | First-person view | `POV walk` |
| Specials | Tilt-shift | ✚ `tilt-shift glide` |
| Specials | Infinite zoom | ✚ `infinite zoom` |
| Specials | Earth zoom out | ✚ `earth zoom out` |
| Specials | Time-lapse | ✚ `time-lapse` |
| Specials | Pass-through objects | ✚ `pass through` (glass: `through glass`) |

**The site's wording is not what compiles.** Its descriptions have no end state
and no negative space — "pan right. Movement: rotate the camera horizontally
from left to right from one fixed point" — which is exactly what the two laws
above exist to prevent. What was taken from the site is the *set*: which moves a
director expects to find. The words are this file's.

Five presets here are on no public list and stay: `dolly counter`,
`mirror/insert POV`, `robot arm`, `drop reveal` and `rack focus`.

### Older ids still resolve

These four surfaces each had their own shorter list before the merge, and their
ids were not the same. A payload written against any of them still lands on the
right preset:

| Old id | Where from | Resolves to |
|---|---|---|
| `locked` | film-setup world default | `locked off` |
| `push` | film-setup world default | `slow push-in` |
| `hand` | film-setup world default | `handheld settle` |
| `track` | film-setup world default | `tracking lateral` |
| `crane` | film-setup world default | `crane up` |
| `rack` | film-setup world default | `rack focus` |
| `pull back` | shot list · review | `slow pull-back` |
| `handheld` | shot list · review | `handheld settle` |
| `tracking` | shot list · review | `tracking lateral` |
| `pan` | shot list · review | `slow pan` |

An alias is read, never written: what goes back into a payload is always the
canonical id.

### Handbook terms resolve too (Zero Shoots, *Camera Work*, 2026-09-08)

The handbook names ten camera movements. Each lands on a preset here; two presets
(✚) were added so that none is approximated:

| Handbook term | Resolves to |
|---|---|
| Circles to the Right / Left | `orbit`, direction in the placeholder |
| Circles Up / Down | `crane up` · `crane down` |
| Camera Rolls Clockwise / Counterclockwise | ✚ `roll` |
| Camera Tilts Down / Up | `slow tilt` |
| Camera Pans Right / Left | `slow pan` |
| Camera Pushes In | `slow push-in` |
| Camera Pulls Back | `slow pull-back` |
| Handheld Shaky Camera | ✚ `handheld shaky` |
| Static Shot | `locked off` |
| Camera tracks the subject | `tracking lateral` — `follow behind` or `front tracking` when the direction is known |

## Getting a model to actually do the move

Every rule below was paid for. Twenty-three of these moves were generated on
`seedance-2.0` from the fragments in this file, measured frame against frame,
and written up in `genre-previews/_camera/` with the prompt, the verdict and the
defect of each one. What follows is what the failures taught, and it applies
whenever you are writing the Camera block — not only when making previews.

### Name the ARRIVAL, never the quantity

The single biggest finding. `orbit` used to read "through roughly 90°" and
delivered about a fifth of that. Rewritten to end on *"their front, their aspect
visibly turned"* it delivered the whole quarter turn, same model, same shot.

A number is an abstraction a model cannot render. A picture is not. "90 degrees"
means nothing; "her face square to the lens" is a frame it can make. So the last
clause of a Camera block is a **destination described as an image** — and this
file's first law, that every move ends somewhere named, is not a style
preference. It is the difference between getting the move and getting a fifth of
it.

### Two kinds of move, two kinds of prompt

| The move's identity is | Write it as | Examples |
|---|---|---|
| an **arrival** — you got there or you did not | our fragment, ending on the destination as an image | orbit, tilt, crane, pedestal, push-in, pull-back, punch-in, drop reveal, zooms |
| a **constant** — something is held for the whole shot | `Movement / Speed / Framing / End` slots, with the constant in Framing | tracking, front tracking, follow behind, chase, handheld, vehicle track, snorricam, drone track, time-lapse |

Confirmed three times: the constant-moves all landed first time on the slotted
form, and the arrival-moves all needed the destination spelled out. Getting this
backwards is how `orbit` under-rotated — the slotted form's End slot says
"complete the intended arc **or full circle**", which is not a destination at
all.

### State what must not move

A model invents geometry unless told not to. Eleven clips carrying this clause
produced no hallucination; the one generated without it grew blue racking on
every wall of a room that had none.

> THE ROOM IS RIGID AND MUST STAY PUT: the benches, the shelving and the window
> wall keep their exact shape and position, and nothing appears or disappears.

Name the actual objects. "Keep it consistent" does nothing.

### The camera does the turning, not the actor

The cheap way to satisfy "we end on her face" is to rotate *her*. That reads as
the move in a single frame and is a lie in motion. Any move whose arrival is an
aspect change needs the guard: **they never rotate, never step, never turn
toward the lens.**

### Camera properties get smoothed; the ACTION slot does not

`handheld settle` was asked for handheld and returned "small organic drift".
`POV walk` was asked for head-height sway with each footfall and returned a
perfectly smooth dolly — zero vertical direction changes across the whole clip,
with the hands never moving at all. Both times the motion had been written as a
property of the camera, and this model irons those flat.

What it animates reliably is what a body does. So when a move's identity lives
in a small repeating motion, give the motion to a body in the ACTION line — arms
swinging and exchanging, a head turning — rather than asking the camera to shake.

**The camera does not then inherit it, and that is the honest limit.** On the
clip that finally worked, the arms swing through the whole shot and the camera's
own vertical trace is still flat. The body carries the rhythm and the tile reads
as walking; the lens keeps gliding. Three framings and three sets of wording
produced no footfall bob at all, so treat handheld sway as something this model
will not give you, and get the motion into the frame some other way.

### What a model cannot be asked to distinguish

It renders the RESULT and is indifferent to the MECHANISM. These came back as
one move each, however carefully the fragment separated them:

- optical zoom against dolly-in — both just "the frame got tighter"
- pivot against short truck — both just "the frame swung right"
- arc against lateral travel — which is why `arc` is retired

So do not spend a clause defending a distinction the picture cannot carry.
Where two presets differ only in mechanism, the honest move is to keep one —
`truck` and `arc` both went for this reason.

### If you are generating a start frame too

- **The start frame pins the camera height**, and no prompt moves it mid-clip.
  A low tracking shot must START low; asking the camera to descend does nothing.
  Same for a tilt: up starts on the feet, down starts on the face.
- **The start frame also decides which motions are AVAILABLE.** A corridor of
  straight facades running to a vanishing point offers exactly one axis — down
  the street — so `snorricam`'s "the whole world lurches behind them" compiled
  into a dolly in and out, which is the one thing a snorricam never does. Reshot
  on a crowded dance floor, with no straight wall and no vanishing point to
  travel along, the same sentence produced the swing: measured, the left and
  right thirds of frame moved the same direction on 96 of 96 steps, where a dolly
  makes them diverge. Before writing the move, ask what the geometry in the frame
  makes easy — the model will do that, whatever the words say.
- **An empty region is not neutral, it is an invitation to invent.** A flat far
  field gets rebuilt frame to frame — a distant pole oscillated left and right
  across four frames of an otherwise clean follow shot. Fill the frame with
  rigid structure.
- **Many small parallax cues beat one big foreground object.** A lone near
  object sweeps the whole frame on any lateral or orbital move and takes the
  subject with it.
- **Dark subject on a pale ground** reads at tile size; pale on pale does not.

### What goes wrong per move, and what to fall back to (2026-09-11)

**Weaker evidence than everything above it, and marked so.** The rules above
were measured on our own clips. The rows below are adapted from the Skool
course's two camera banks — a thirty-move *prompt library* carrying a *risks*
and a *safer version* field, and a thirty-seven-move *full catalogue* carrying
an *AI note* per move, the model's characteristic failure for that path. One
practitioner's experience stands behind them, not a measurement of ours. Treat a
row as a thing to check before dispatch, not as a finding
(`research/ai-video-bootcamp/ANALYSIS.md` §4.4).

**Read the third column as an instruction to the prompt, not to the shot.** Most
of these failures are the model filling a slot we left empty. Where our fragment
already fills it, the row says so — those are worth keeping precisely because a
later edit that "simplifies" the fragment would reopen the hole.

| Preset | Goes wrong when | Fall back to |
|---|---|---|
| `locked off` | anything short of refusing movement outright gets a slow drift added | our fragment already refuses it in as many words — the row exists so that clause is never trimmed |
| `slow pan` | the pan has no destination, so it becomes the same aimless drift a missing instruction produces | name what the pan lands on, not only that it ends composed |
| `slow tilt` | a tilt mixed with any pan warps geometry, worst on a face | one axis only, and name both ends — where it starts, where it lands |
| `slow push-in` | nothing in the scene triggers the move, so it reads as aimless drift; held too long, it stops reading as real time | a short straight push that begins on a visible trigger and ends on a named held frame |
| `slow pull-back` | the thing revealed is vague, or the space behind the subject is underdescribed, so the reveal arrives empty | open on one clear subject and reveal exactly one new layer |
| `slow zoom in` / `slow zoom out` | — | the safest move that exists on a face: no camera travel means no parallax, so nothing warps. Reach for it when a push-in worries you |
| `fast zoom in` / `fast zoom out` | the faster the zoom, the harder the model drags everything else into slow motion | the real-time clause in the shot header is load-bearing here, not decoration |
| `crash zoom` | the snap lands on nothing in particular, so it reads as a glitch rather than emphasis | land it on a named feature — the eyes, the hands, one object |
| `crash zoom out` | the wide arrives but the new information was never described, so the shock has nothing to be about | say what the wide reveals; the reveal *is* the shot |
| `slider` | nothing real sits close to the lens, so there is no parallax and the move degrades into a small pan | put a named object in the near foreground — the parallax is the whole point of the move |
| `pedestal` | the model blends the rise with a tilt into a curve | our fragment already says lens level, no tilt |
| `push through` / `through glass` / `pass through` | the opening is only half crossed — the commonest failure of this whole family | the barrier must be fully passed, and the space beyond it named |
| `orbit` | a full circle drifts both identity and room geometry; and if nothing in the frame changes during the move, the orbit is decoration | a part turn ending on a named aspect — our arrival rule above already fixes the under-rotation half of this |
| `follow behind` | — | tracking legitimately begins already in motion, so its start needs no instruction. Only the ending is still yours to decide, and it still needs deciding |
| `front tracking` | — | the most stable tracking there is for a face: the subject barely moves relative to the lens, so identity holds |
| `tracking lateral` | the background was never described, and the background is where the sense of speed actually lives | name the layers that will stream past behind them |
| `low tracking` | the face creeps into frame because nothing forbade it | "the face is never shown" is an instruction models respect — say it rather than trusting ankle height to imply it |
| `vehicle track` | the camera is not anchored to anything, so the shot floats free of the vehicle | name what it rides — a parallel car, a rig on the door |
| `chase` / `handheld run` | the shake is constant and uncaused, which reads as an effect rather than an operator | every correction needs a cause: a turn, an obstacle, a stumble |
| `handheld shaky` / `handheld settle` | asked for as abstract "shakiness", which models render worse than a described human | our fragments already name the operator's breath and weight — that is the form that works |
| `snorricam` | the rig is not described as rigid, and it returns as an ordinary handheld shot of a face | keep "rigged to the body" explicit; our fragment does |
| `crane up` | the high frame was never designed, so the rise lands on nothing; too fast a rise throws away the intimacy it opened with | start close enough to establish the person, end on one clear wide payoff — and say the subject stays in frame as the camera climbs, or it becomes a fly-away |
| `crane down` | the descent has no named landing, so it drifts to a stop | name the one window, one face, one object it arrives on |
| `drone establish` / `drone track` | the flight has no line to follow, so the drone wanders | give it a road, a river, a coastline, a wall |
| `whip pan` | used as a hidden cut with the blur on either side unmatched in direction, density or brightness — then the cut shows; busy content inside the blur reads as noise | match direction and blur on both sides, and keep the blurred frames empty |
| `pass-by reveal` | the camera leaves the subject without arriving at something that matters, so the move reads arbitrary | make the destination clearly more important than what was left behind |
| `roll` | the space has no strong geometry to rotate against, so there is nothing to read the roll against; past a point the chaos is unreadable | keep firm horizontals and verticals in frame as the reference |
| `rack focus` | focus and camera movement begin at the same instant, muddying the cue; or the first object was never clearly established | land the focus change first, then move — and our fragment already holds the frame still through it |
| `infinite zoom` | more than about two worlds in one clip and the geometry stops holding | two per clip; chain clips for a longer dive |

**Where the source is wrong for us, and we keep ours.** Three cases, all worth
recording so the conflict is not rediscovered:

- **Zoom against dolly.** The catalogue says naming *"the camera stays in a
  fixed position"* stops the model blending them. Our fragments have said
  exactly that for some time, and our own probe still found the two coming back
  indistinguishable — the model renders the RESULT and is indifferent to the
  MECHANISM (above). His fix is already applied here and measured insufficient.
  Keep the clause, keep the expectation low.
- **`orbit`.** The library's fallback is *use fewer degrees*, which is the
  quantity form this file's first law rejects. The arrival form is the fix we
  measured.
- **Dolly zoom.** The catalogue gives an anchor line for forcing it — hold the
  subject's size exactly constant while the background moves. We **retired**
  dolly zoom by decision; it resolves to `slow pull-back`. This note is recorded
  as the thing to try *if* that decision is ever reopened, and is not a reason to
  reopen it.

### Big face, simple move (2026-09-11)

**The tighter the frame on a face, the simpler the move has to be.** A wide
shot will absorb almost anything — a sweep, a rise, a circle — because no single
part of the frame is under enough scrutiny for a small error to show. Once a
face fills the frame, the same move has to hold a likeness steady through every
one of its frames, and that is where identity drifts and features warp.

So at close-up and tighter, stay inside: `locked off`, `handheld settle`, a
short `slow push-in`, `slow zoom in`, `rack focus`. Save `orbit`, `crane up`,
`roll`, `whip pan`, `pass through` and the aerials for the wides.

This is the same finding as `orbit`'s row above — *the wider the shot, the safer
the circle* — stated once as a rule rather than once per preset, because it
holds for every path in the kinetic register and not only the circular ones.
Adapted from the course's camera lesson, and **not measured here** — no probe of
ours has tested move complexity against shot size.

## Shot-size ladder — a DISTANCE, and nothing else

Nine rungs, wide to tight — the handbook's seven plus our `MCU` and `MWS`
(2026-09-08, the user's call: the board shows every size the book has). A size
says how much of the figure the frame holds
and **nothing about who else is in it or where they stand** — that is the
arrangement below, and keeping the two apart is what stops "OTS" being used as a
distance.

**Ids:** `detail` · `ECU` · `CU` · `MCU` · `MS` · `MWS` · `WS` · `EWS` · `establishing`

*(The sentence each id compiles to now lives server-side. Pass the id to `picsart_film_compile_prompt` and it pastes the tuned wording; never write that sentence yourself.)*

**The handbook's seven shot sizes are all rungs (2026-09-08).** Establishing
Shot → `establishing`, Extreme Wide Shot → `EWS` (its sentence is the fragment),
Wide Shot → `WS`, Medium Shot → `MS`, Close Up → `CU`, Extreme Close Up → `ECU`,
Detail Shot → `detail`. Two of them lean on the arrangement beside them: a
`detail` is normally shot as `insert` (a thing, no face — a hand may enter), and
an `establishing` opens the scene on its `single` or `group`, first in the
scene, before the action. The size still says only how much the frame holds;
the arrangement says who is in it.

## Arrangement — who is in frame, and how they sit in it

**`OTS` and `INSERT` used to be rungs on the ladder above.** They are not
distances: an over-the-shoulder can be a medium or a close-up, and an insert has
no figure to measure at all. Worse, a size chosen from a single list quietly
decided the staging too — a plan asking for "OTS" was making a claim about two
bodies holding fixed positions, and nothing checked it against the scene's map.

So arrangement is its own decision, and **anything with a second body is a claim
that gets checked** before a shot renders.

**Ids:** `single` · `two-shot` · `dirty single` · `OTS` · `over the hip` · `triangle 2+1` · `triangle 1+2` · `face to face` · `back to back` · `POV` · `insert` · `group`

*(The sentence each id compiles to now lives server-side. Pass the id to `picsart_film_compile_prompt` and it pastes the tuned wording; never write that sentence yourself.)*

`insert` pairs with any tight size; `single` is the default and the only one that
makes no claim about a second person. The five rows from `over the hip` to `back
to back` are the handbook's *Character Blocking* (2026-09-08), its wording kept;
`<subject's character view>` is filled from the Character view table below —
the handbook writes the far subject's facing into the blocking line, and so do
we. Anything with a second body is still a claim checked against the scene's
map before it renders.

## FOV ladder — ten anchors, one per shot, locked

FOV changes only on a hard cut. Every shot's degrees go into its prompt — pass
the card's lens as the shot's `fov` to `picsart_film_compile_prompt`, whether the user
changed it or left it as seeded (`PROMPT-BLOCKS.md`, Part 2, *The card's own
picks*, revised 2026-09-30); the lens equivalent is for talking to humans.

| Anchor | ≈ lens (ff) |
| --- | --- |
| 10° | 200mm |
| 15° | 135mm |
| 24° | 85mm |
| 31° | 65mm |
| 40° | 50mm |
| 54° | 35mm |
| 65° | 28mm |
| 74° | 24mm |
| 90° | 18mm |
| 104° | 14mm |

*(The sentence each id compiles to now lives server-side. Pass the id to `picsart_film_compile_prompt` and it pastes the tuned wording; never write that sentence yourself.)*

**These degrees are the real ones.** The shot list used to carry a six-rung
version of this same ladder with four of its anchors wrong — 50mm labelled 36°,
135mm labelled 14°, 24mm 75°, 18mm 94° — so the same lens had two different
numbers depending on which surface you were standing on. A full-frame 50mm is
39.6° horizontal; the anchors above round it honestly. The retired ids resolve
(see below) so an approved plan still lands on its lens.

**This ladder is the film's only focal length** (2026-09-21). The setup's
Focal length pill is gone; its Lens pill is the glass *family* (spherical,
anamorphic, vintage…), which says how the picture renders, never how wide it
is. A setup locked before then may still carry `focal` in millimetres (14 · 24 ·
35 · 50 · 85 · 135) — its stored prefix keeps it, and the cards decide from here
on. Aperture moved the same day: the setup has none, and the table below is the
only one. That is also why a seeded lens is no longer skipped: with no focal
length in the style line, a shot whose lens was not passed would reach the model
with no focal length at all.

## Aperture / depth

**Ids:** `f/1.4` · `f/4` · `f/11`

*(The sentence each id compiles to now lives server-side. Pass the id to `picsart_film_compile_prompt` and it pastes the tuned wording; never write that sentence yourself.)*

## Angle / height

**Ids:** `eye level` · `low` · `high` · `overhead` · `drone top-down` · `top-down foreshortening` · `low foreshortening` · `three-quarter` · `dutch`

*(The sentence each id compiles to now lives server-side. Pass the id to `picsart_film_compile_prompt` and it pastes the tuned wording; never write that sentence yourself.)*

The `low`, `high` and `dutch` fragments took the handbook's wording on
2026-09-08, and the three new rows are its *Camera Angle* table — the two
foreshortening shots (its low one is written as a five-line spec; the fragment
above is that spec as one clause) and the drone top-down. Its *Eye Level*, *POV
shot* and *Dutch* were already here; POV lives in Arrangement, since it is a
claim about who is in frame.

## Character view — which way the subject faces the lens (2026-09-08)

The handbook's axis, and one these tables lacked: the camera's height says
nothing about whether we see a face or a back. Written into the blocking line for every
present character; the handbook's phrases are the fragments.

| Preset | Use it when |
| --- | --- |
| front | the face matters and the audience must read them |
| profile | direction, silhouette, or face-to-face blocking matters |
| three-quarter | both face and side angle visible — the cinematic default |
| three-quarter back | leaving, looking over a shoulder, a partial reveal |
| rear | the face must not be seen; the back and the body's direction matter |

*(The sentence each id compiles to now lives server-side. Pass the id to `picsart_film_compile_prompt` and it pastes the tuned wording; never write that sentence yourself.)*

## Placement — where the subject sits in the frame (2026-09-08)

The handbook's three terms. Centered is direct and confrontational; a third opens
negative space, creates direction and leaves room for the second subject or the
environment. the blocking line, one per shot when it matters.

**Ids:** `centered` · `left third` · `right third`

*(The sentence each id compiles to now lives server-side. Pass the id to `picsart_film_compile_prompt` and it pastes the tuned wording; never write that sentence yourself.)*

**Which third is not a free choice — leave the room in front (2026-09-11).**
The open side goes where the subject is looking or moving: someone looking
frame-right sits on the *left* third, so the space they look into is in front
of them. Put the empty half behind them instead and the shot reads as trapped,
about to walk into the edge — the frame is composed correctly and still feels
wrong, which is why this one is hard to spot in review. Say it in the blocking
line as the space, not as a rule: *"Rasa on the left third, the length of the
platform open in front of her."* Ours, adapted from the Skool course's framing
lesson; not measured here (`research/ai-video-bootcamp/ANALYSIS.md` §4.11).

Not to be confused with the eyeline rule in `SEAMS.md`, which is about two
shots across a cut — A looks off frame-right, and the thing is on frame-left in
B. Lead room is about the space inside one frame. The two are independent:
obeying one says nothing about the other.

## Composition — how the frame is organised (2026-09-08)

The handbook's five, its wording kept. the blocking line, only when the frame's
organisation carries meaning; a shot that names none is composed by the
arrangement and the placement alone.

| Preset | Good for | Use it when |
| --- | --- | --- |
| symmetrical | formal scenes, controlled environments, a centered subject, balanced power | balance, order, ritual, control, a clean centered frame |
| leading lines | hallways, roads, tables, lights, ceiling lines, counters | the environment should guide the eye to the subject |
| separated by distance | emotional distance, threat across the room, foreground/background contrast | distance should create tension, loneliness, power or scale |
| side by side | partnership, comparison, walking together, matching reactions | two subjects share the frame on one plane |
| vanishing point | roads, hallways, long rooms, aisles, streets | the scene should have depth lines pulling toward one point |

*(The sentence each id compiles to now lives server-side. Pass the id to `picsart_film_compile_prompt` and it pastes the tuned wording; never write that sentence yourself.)*

## Atmosphere — what is in the air between the camera and the distance (2026-09-11)

Ours, not the handbook's. The second tool for the job the three planes do: the
planes give the eye somewhere to travel, and atmosphere is what physically
separates them. Air is not empty, so the further away a thing is the more air
sits in front of it — distant things lighten and lose contrast, and the eye
reads that as distance before it reads anything else. Naming it is the fastest
way to stop a frame reading as a subject pasted on a backdrop.

| Preset | Use it when |
| --- | --- |
| haze | a big exterior should feel vast; the background must sit far back |
| low fog | the planes must be unmistakably separate; ground-level menace or cold |
| light shaft | an interior with one strong source; the air should be visible |
| smoke drift | movement and mood on top of depth — a fire, exhaust, a kitchen |

*(The sentence each id compiles to now lives server-side. Pass the id to `picsart_film_compile_prompt` and it pastes the tuned wording; never write that sentence yourself.)*

**It belongs to the place, not the shot.** A street that is hazy in one picture
is hazy in all of them, or the location comes apart worse than a moved wall
would break it. So it is decided at the location's plate, carried in the
`Setting` line of every picture made there, and never introduced by a single
shot — *a frame outranks the words* (the language laws), so a shot asking for
fog the plate does not have will fight its own reference and usually lose.

Check it against the locked look before using it: a stock with strong halation
and a haze clause will wash the far plane twice. Adapted from the Skool course's
framing lesson and not measured here
(`research/ai-video-bootcamp/ANALYSIS.md` §4.11).

## Camera artefacts — the small failures that prove a camera was there (2026-09-11)

A real camera is a physical object with a lens that has to hunt for focus, a
meter that has to catch up to a changing light, and a front element that things
land on. A generated shot has none of those, and its *absence* is one of the
tells: the image is impossibly clean and impossibly decided, and it reads as a
render for the same reason a face with no pores does.

This is the moving half of the family the video prompt's quality line already
opens with frame rate and shutter angle (`PROMPT-BLOCKS.md`, Part 1 item 7).
Those two say how motion blurs; these say what the instrument does while it
films.

**They ride the quality line, and never the Breathing feel line.** Breathing
feel is one of the handbook's two camera lines carried *verbatim* (Part 1 item
5) — it is not ours to extend, and appending to it is exactly the failure the
book-fidelity rule exists to prevent. The quality line in item 7 is marked ours
and already carries the skin clause and the frame rate; that is where these
belong. Never into `stylePrefix` either — like shutter angle they are **video
only**, and a still has no focus hunt.

**Ids:** `focus hunt` · `exposure shift` · `lens contact` · `late reframe`

*(The sentence each id compiles to now lives server-side. Pass the id to `picsart_film_compile_prompt` and it pastes the tuned wording; never write that sentence yourself.)*

**One per shot at the very most, and not in most shots.** These read as an
accident, and an accident that happens on every shot is a style — at which point
it stops proving a camera and starts proving an effect. The course's own warning
about handheld shake applies unchanged: the moment the viewer notices it as a
technique, it has failed.

**Check against the film's chosen texture first.** A film whose look is formal —
locked-off frames, controlled light, the polished registers — is actively hurt by
a focus hunt, the same way a haze clause hurts a stock with strong halation. The
fragments belong to a film that is pretending to be found or observed, and the
Breathing feel line is where that is already decided.

**Weaker evidence than the rules above.** Adapted from the course's prompt
banks, where these appear inside worked prompts rather than as a rule; not
measured here. The related move, a handheld camera snapping into a subject and
correcting — what the course calls the documentary snap zoom and rates as its
single strongest realism trick — has **no preset here**, deliberately: it is two
moves in one shot, which the one-move-per-shot law refuses. If it is ever wanted
it is a shot design, not a fragment.

## Retired ids that still resolve

Every surface had its own shortened version of the tables above before they were
merged. A payload written against any of them lands on the right preset; an alias
is read and never written back.

| Old id | Where from | Resolves to |
|---|---|---|
| `truck` | retired preset | `tracking lateral` |
| `lead walk` | renamed | `front tracking` |
| `arc` | retired preset | `orbit` |
| `dolly zoom` | retired preset | `slow pull-back` |
| `top-down` | shot list | `overhead` |
| `14°` | shot list | `15°` |
| `36°` | shot list | `40°` |
| `75°` | shot list | `74°` |
| `94°` | shot list | `90°` |
| `wide` | film-setup framing | `WS` |
| `medium` | film-setup framing | `MS` |
| `close` | film-setup framing | `CU` |
| `xclose` | film-setup framing | `ECU` |
| `ots` | film-setup framing | **dropped** — an arrangement, not a size |

## Without boards

Offer movement as one question (4 options drawn from the scene's energy, preview =
the fragment), size + FOV + aperture as a second, angle folded into whichever
needs it. Character view, placement and composition are written from the
staging, never asked. Record chosen preset names in the shot's `design`; paste
fragments into the blocking and camera lines unchanged.