Skip to content

The blocks

Six blocks. Fill them in any order; leave any of them empty. A brief with a good Intent and nothing else is more useful than a brief with all six filled badly.

The grade, palette and film language everything in the project is rendered in.

Unlike the other five, the Look is chosen, not written. Four groups, each picked from a catalog: Format (frame rate, shutter, aspect), Lighting, Emulsion (the stock and its grain), and Color (a grade plus a palette of three named colors in dominant, secondary and accent slots). Frame and lens are per-shot and live in the shot, not here. You pick from the catalog’s vocabulary rather than describing these in prose, because the models respond far better to that vocabulary than to yours or mine.

That is also why the Look is the one block a document upload does not fill. The Brief Agent advises on it in prose, and writes it in one case only: attach a reference picture to the Look block and ask it to propose from that, and its Lighting, Emulsion, Grade and palette arrive as a proposal you apply or discard. See Reference images. The choice stays yours either way.

Good content: a coherent set. A stock, a grade and a palette that belong to the same film. Picking a warm grade and a cold palette gets you a fight rather than a compromise.

A single shot can override the project’s Look where it needs to. The brief sets what everything inherits.

What this is, who it is for, where it runs, and the tone. Five fields:

Field What goes in it
What this is The piece, in a sentence. “A sixty-second spot for a fishing co-operative.”
Audience Who watches it, specifically. Not “everyone”.
Where it runs The surface. Phone, cinema, trade-show wall, a feed with sound off.
Tone How it should feel. “Plain, unsentimental.”
Notes Production-wide rules. “No drone shots. No logos larger than a chest patch.”

Intent does more work than any other block, because it is what the Prompt Agent consults before it writes anything at all. A good “where it runs” changes framing decisions in every shot without you ever mentioning framing.

Two things about format ride with Intent without being Intent fields. The Brief Agent asks once, in the Intent round, for the frame rate and the shape of the picture; and an uploaded document that states a frame rate, shutter or aspect shows them on the Intent draft as a Format row. Confirming Intent writes those to the Look’s Format and touches nothing else in the Look. A shot you open in the Studio then starts on that aspect. Change it on one shot and that shot keeps your choice.

The project’s models ride the same way. A document that names them (“Motion on Seedance 2.5. Stills on Nano Banana Pro”) shows them on the Intent draft as a Models row, one cell per mode; a mode the document does not name shows Lorvara’s default, marked as the default; a model Lorvara does not offer leaves its cell empty for you to choose. Confirming Intent sets the models of the project you opened the brief from: stated in the brief, stored on the project, so the same brief can serve two projects on different models. The Brief Agent asks for them in the Intent round too.

The names to write are the ones on Lorvara’s current model list, https://lorvara.com/api/models.json. It is read live from Lorvara’s own records, so it is never out of date, and it is plain JSON an assistant can read. If an assistant is writing your brief, tell it to fetch that list before it fills the Models line, and to use the names exactly as the list writes them.

In a document you upload, Intent is the one block with no heading of its own. Its lines sit under the title, and Lorvara reads them from the whole document rather than from a section, so do not write ## Intent. Cast, Locations, Props and Glossary each have their ## heading.

Good content: concrete and restrictive. “People in coastal towns who already know the boats” tells the agent something. “General audience” tells it nothing and it will write as though the field were empty.

Notes are for production-wide rules only. A blocking rule that applies to one location belongs in that location. A costume rule that applies to one character belongs on that character. Keeping them where they belong is what lets each block be read on its own.

There are length limits: 300 characters for what this is, 200 each for audience, where it runs and tone, 600 for notes. They are deliberate: these fields are read on every prompt, and a long one crowds out the shot you are actually writing. If you are over the limit, the answer is almost always that something in there belongs in another block.

The named people. One card per character.

Each card carries:

  • A name. The one you will refer to them by.
  • A description. How they look, in enough detail that a model renders the same person twice.
  • States. The versions of them that matter, each with a name and what changes. “In oilskin, on the boat.” “In an evening coat, ashore.” A character with no states is fine. Most are.
  • A default state, if they have several.
  • Reference pictures, per state: up to five views (front, three-quarter, face, profile, back) and a four-view reference sheet. A state can hold both. A shot takes one picture per reference it names: each angle it names is one picture, so naming the sheet and the face sends both. See What a shot receives.
  • Voice: see below.

Good content: specifics that survive rendering. Height, build, age, hair, the one distinguishing thing. “Weathered” is not a description; “sixty, sun- damaged, a white beard cut close, a nose broken and badly set” is.

States earn their place when something changes on camera. If the character wears one thing throughout, do not invent states for the sake of it.

Voice is not a block. It is split across two, because its two halves belong to different things.

How a character speaks lives on that character’s Cast card, as delivery notes: pace, register, accent, and free notes. It belongs to the person, so it travels with them wherever their card goes.

How a word is pronounced lives in the Glossary. It belongs to the word, not to whoever happens to say it: a place name is said the same way by everybody in the production, and putting it on one character’s card would lose it the moment anybody else spoke.

So: “Tomas speaks slowly, drops the ends of his sentences” is a Cast delivery note. “Tomas is toh-MAHS, not TOH-mas” is a Glossary entry.

If you only ever remember one thing here: people go in Cast, words go in Glossary.

Leave a delivery field out if it was never established. An empty accent field is a true statement; a guessed one is a lie your prompts will repeat.

The places. One entry per place.

Each carries:

  • A name.
  • An anchor: the one unmistakable feature that identifies this place and no other. The green light at the end of the mole. The third mooring ring with the frayed blue rope. The anchor is the most important field in the block and the one people skip.
  • A description of the wider place.
  • Never from: the camera positions this place rules out, ticked from a catalog: directly above, from behind, and so on. Usually none. This is a list of restrictions, not recommendations: a position the brief happens to like is not a rule against the others, so it is left unticked.
  • Reference plates: up to two, each with a label.

Good content: the anchor does the work. Without one, “the harbour” is any harbour and every shot gets a different one. With one, every shot set there is demonstrably the same place.

A note on camera positions: the camera is otherwise the shot’s to choose. A shot can ask for anything the place does not rule out. The location saying “never from directly above” is a default that saves you writing it forty times.

The things that matter, and the states they are in. Structurally the same as Cast (an entry with a description, states, and reference pictures) but for objects rather than people.

A prop earns an entry when it recurs and its condition changes: the trawler tied up, leaving, and being hosed down; the lantern lit and unlit. A prop that appears once and never changes is usually better described in the shot.

Good content: states that are visibly different. “New” and “slightly less new” is not a pair of states. “Dry” and “soaked through” is.

Names and terms, with how they are said and what they mean.

Each entry has:

  • The term: a name, a place, a registration number, a piece of jargon.
  • How to say it: a plain-English pronunciation. Write it the way you would tell someone out loud: toh-MAHS, why one four one. There is no phonetic alphabet to learn.
  • A note: what it is, if that is not obvious.

Good content: anything a stranger would get wrong. Proper nouns, boat registrations, local place names, invented words, initialisms that are said as words rather than spelled out.

Glossary is small and cheap and it fixes a class of problem (a name said four different ways across a piece) that is otherwise very tedious to catch.