Skip to content

How do Scenes & Shots work?

A shot list from parse to confirm, and the prompts it produces

The brief describes the world. The shot list says what you are making today, in it. One card per shot, with its description, mode, duration, cuts and references, reviewed, confirmed, and handed over as a batch of work.

The description is the card. It is full width and grows as you type, because it is the line every composition re-reads. Everything else on the card sits around it.

Under it, in one row: Mode (image or video), Duration, Cuts (single or multi-cut), and the References this shot names, typed as a comma-separated list.

The reference chips read back what that list resolves to. Beneath the field, one chip per name:

  • green, with a tick — an image is on file and it will be attached when this shot composes;
  • amber — the name is a brief entry with no image on file, so the shot is composed from its words alone;
  • plain — a name Lorvara cannot place, typed but not on the brief.

The chips are a reading of the field, not a second control. The field is still the one thing that saves.

In the card’s header: its number within the scene, its name, a status select, and a ⋯ menu.

The status is yours alone: Working or Complete. Nothing else reads it: not Confirm, not composing, not the lock. It is a marker for you in a long list.

The ⋯ menu holds the three acts that do not earn permanent space:

Item What it does
Move to scene… Moves this shot to another scene, or out of every scene.
Dialogue Opens a Studio speech prompt seeded from this shot: its text, its duration, and the cast voice and glossary terms the text names.
Remove shot Takes the row off the list.

Drag to reorder within a scene. A shot’s position is its place among the shots of its scene, so reordering renumbers cleanly. Dragging across scenes is not how you move a shot. That is Move to scene, because it changes what the shot is composed against.

Guesses are marked. Where the parser inferred a mode, duration, cuts or references rather than reading them, the cell is dashed and marked guessed. Editing it clears the mark; Accept guesses at the foot of the list accepts mode, duration and cuts as they stand. References have their own act, I intend to attach these, because promising a reference tells the agent those images will be there.

Each model takes a limited number of reference images, and the project’s default model for the shot’s mode decides which limit applies. When a shot names more pictures than that, the ones you named first attach and the rest are held. Lorvara walks the References in the order you typed them and attaches each while it fits; the first that does not fit is held, and so is everything named after it. The limit counts pictures, so a location counts once for each plate it brings.

For instance, name six on a model that carries four, and the first four you named attach; the last two are held.

A held reference sends no picture, and the composed prompt does not address it: there is no @ token and no instruction for it, and the agent is told it was held and why. Your References field is left exactly as you typed it. Lorvara does not remove anything from it for you.

Before you compose, the card shows an i next to the field. Its note reads:

Names more than <model> carries: the first <n> attach; not sent: <names>. Trim the References to <limit> or change the project’s default model.

After it composes, the card carries what that composition reported:

When last composed: <model> attaches at most <limit> reference images; this shot names <count> — the first <n> attach; not sent: <name> (<@token>), …. Trim the shot’s References to <limit> and recompose, or change the project’s default model

Either way the fix is yours: take references out of the field and recompose, or point the project at a model that carries more. On a model that takes no reference images at all, none attach, and the note says so: “Names more than <model> carries, none attach.” Which picture each name sends is in Reference images.

Some rows are not shots at all. A reference row asks for a picture to be made: a character or prop in a named state and one view, or its whole reference sheet, or a location plate. What it produces is the input later shots stand on.

A reference card looks different on purpose. Its words are the brief entry’s own, shown read-only, because that is where they are edited. Open Brief on the card goes straight there. Its mode, duration, cuts and references are fixed and shown as such: Image, —, Single, none — it is the reference. The one act it carries is Remove; there is no ⋯ menu, because Move to scene and Dialogue do not apply to a reference.

Reference cards live together under Named references, below the scenes.

+ Add shot sits at the end of each scene. It adds an empty row to that scene.

+ Add scene sits after the scenes, so the button is at the end of the thing it adds to.

+ Add reference sits at the end of Named references, and it opens a picker rather than an empty row. A reference’s words come from its brief entry, so there is nothing to type. The picker lists what the brief holds, grouped into Characters, Props and Locations, with the states and views each one offers.

The picker says its price before you press. An option already on the list is marked with a ✓ and picking it goes to the row that exists, creating nothing and spending nothing. Anything else is created and composed on the press, charged at the same rate as any composed row.

Below the scenes you may see a violet panel headed Other references, saying how many brief entries no shot names.

That is a completeness check, not a suggestion: a shot names an entry with a Reference:, an @token, or a reference row, and these entries have none. A character who is only ever “the woman at the bench” is listed. Tick the ones you want and press Add reference rows; nothing is added until you do.

It is deliberately not the way to get a reference for a character your shots do name. That is Create shot prompt on the brief.

Confirming is the hand-off, and it does not generate anything. The button reads Confirm & compose and says how many prompts it will make, in how many scenes, and what composing them costs. It turns every row into a Key Prompt and writes each one’s text, and stops there. Nothing is generated and nothing is spent on pictures or video until you generate, prompt by prompt. Composing the text is itself charged; see Recomposing, and what it costs.

On a list you have already confirmed, the same button becomes Save & re-compose, and it names the number and the price. See the same page.

One list holds up to 40 shots. A longer shoot is worked in passes; see The shot list for how.

A scene can override the Look. The project’s Look is what everything inherits; a scene that needs to depart from it (a flashback, a different time of day) carries its own. A scene with nothing overridden shows its Look folded into one line; open it to set one.

There are exactly two ways a Key Prompt comes into existence, and knowing which one you are using is most of understanding this page.

Confirm & compose on this page turns every row into a Key Prompt and writes its text. That is the bulk path, and it is how a shoot’s shots become work.

On a confirmed brief entry (a character in a state, a prop, a location) there is Create shot prompt. One press makes the reference row here, mints the prompt and composes it. Characters also offer tick boxes and Create N shot prompts, N being what you ticked: the four-panel reference sheet, plus a face view if you tick it.

Press it twice for the same thing and nothing is charged. It tells you the prompt already exists and stops.

That single press refuses in two situations, because it would otherwise do more than it says:

  • The list has rows that are not prompts yet. Confirming would mint and compose all of them, not just yours. Finish your review here first.
  • The brief’s shots have not been sent to this project, and no list is confirmed. The press would confirm the list, and a confirmed list closes the project to those shots, so they would be stranded.

Neither refusal spends anything.

Confirming moves the shots here for good. The brief’s end-of-page line says so plainly:

<Project> has confirmed its shots — they are edited on its Scenes & Shots now. Reference images are still added here, on the brief.

Both halves matter. The shots are yours to edit here, and the brief will refuse to send that list again. The brief is still the only place reference pictures are added, and still the place an entry’s words are edited, so “not from the brief” is about the shot list, never about the brief being done with.

The lock is per project. Another project using the same brief is unaffected and can still take the shots.

Get the brief’s blocks confirmed first. The shot list reads the brief at the moment you confirm, not before. A list confirmed against an empty Cast produces shots that have never heard of your characters, and the fix is to fill the blocks and run it again. Nothing writes back the other way: composing a shot never changes the brief.

A shot’s position is computed, never stored. It is the row’s index among the shot rows of its scene, worked out wherever it is displayed. That is why reordering a scene renumbers it cleanly, and why a reference row sitting between two shots does not renumber them.

The shot’s name is its own field, and this is load-bearing. Titles are composed from the name plus the coordinates, rather than the title being the stored thing, because Shot 3: the reverse is a perfectly legal name. If the two were ever collapsed into one string, the first reorder would either destroy a name or refuse to run.

Once a row has a record, the record owns its name. The card shows the record’s composed title, read-only; rename it in the Studio or the Library.

A shot list belongs to a project, not a brief. Two productions in the same world do not share a shot list. That is what makes them two productions. See The shot list for why the split is the shape of the two things rather than a limitation.