How the chat works

There is one chat in Filmara, and one voice in it. You ask for work in your own words; it answers, or it asks you something back. There's no cast of assistants to pick between and no mode to switch into first.

Working with the AI covers what to ask for. This page is about the mechanics underneath — what the @ menu does, what Filmara can actually see when you ask, and where the stop points are.

Your words come first

Plain sentences work, and they're the fastest route. "Write a scene for the opening beat." "Tighten the third scene." "Who is she, really?" You never have to learn a syntax to get work out of Filmara.

@ is for when you want to be exact — to name the task, the scene, or the character rather than let Filmara infer it. Reach for it when you've asked for something and got the wrong scene, or when you want a specific piece of work without describing it.

Typing @

The menu offers three kinds of thing:

• Tasks — summon a specific piece of work.
• Scenes — point at a scene by its heading.
• Characters — point at someone in your cast.

Scenes and characters behave differently from tasks: picking one attaches it as a chip above the message rather than leaving text in your sentence. The chip is what travels — it carries the actual scene, not a name Filmara has to guess at. Pick two, and both travel.

The tasks you can summon

Type this	And Filmara will
@character	Build a new character, or deepen one that already exists
@environment	Build an environment — the place, its mood, how light works in it
@shot	Break a scene into shots you can render
@direct	Stage a scene — how it plays, moment by moment
@item	Make a world item and file it in your Inventory
@complete	Take a shot that isn't ready and fill in what it's missing

Each one answers with a card you can read before anything is committed. None of them changes your project on its own.

Two pieces of work deliberately live elsewhere:

• Editing a scene — just say it. "Cut the phone call." "Make her colder in the second half." Filmara works out which scene you mean and stages the change as a reviewable diff in the editor. There's no summon to learn.
• Rendering a shot — set up in the Producer lens, with Propose render on the shot itself. A chat message doesn't carry the shot's chosen still, its camera move or its length, and Filmara won't invent them — so if you ask in chat, it'll point you at the shot instead of guessing.

It knows where you are

Every message carries a quiet pointer to where you were standing when you sent it — the scene your cursor is in, the shot you have open, the beat you're looking at. So "break this one into shots" works without naming anything.

If your words clearly point somewhere else, your words win. The pointer is a default, not a cage.

It asks instead of guessing

This is the behaviour worth trusting most: when Filmara doesn't have what a job needs, it asks you for it rather than inventing something plausible.

Ask for shots with no scene in view and you'll get:

I can break a scene into shots once there's a scene to work from. Which scene should I cover? Open one in the editor, or tell me what happens and I'll frame the shots from that.

Summon a character with nothing to go on and it asks who. Ask to render a shot that has no still chosen yet and it tells you to make one on the Storyboard first, then come back.

Every one of these is a real precondition being checked — not a canned apology. A blank profile or a shot list for a scene that doesn't exist is worse than a question.

Nothing changes until you say so

There are three shapes of stop point, and everything the agent produces passes through one of them:

• Cards — a character, a location, a shot list, an item. Read it, then accept or dismiss. Nothing lands in your project until you accept.
• Staged edits — changes to your screenplay arrive as an inline diff in the editor, hunk by hunk. You accept the ones you want.
• Plans — when your ask is a job rather than an answer, you get a plan card listing every step, what it read of your project, and the credit ceiling if any step generates images. Nothing runs until you press Start, and Start authorizes exactly what's on the card.

Anything a plan writes sits above a checkpoint in Versions, so a whole run is one step to undo.

You can see what it read

Above each answer, Filmara shows what it actually had in front of it for that turn — the scene, the brief, your characters, the decisions already settled. When an answer seems to have missed something, that's the first place to look: usually it wasn't in the room.

It remembers this project

The conversation is per project, and it persists. Filmara keeps what you've settled — the premise, who your characters are, decisions you've made about the story — and stays consistent with them rather than re-litigating them every session. Older conversation is compressed rather than dropped, so a long project doesn't lose its early ground.

When it's a job, not an answer

Some asks are too big for a single reply — "create this story end to end" is a structure, a cast, and a scene per beat. Those become tasks: a plan you read first, then run one step at a time, with a progress log you can watch and a pause any time you type.

Working with the AI walks through the asks that work this way, what each produces, and what it costs.