Skip to content
EpicDraft Docs

Using AI Commands

An AI Suggestion answers "what comes next". An AI Command answers "change what is already there": tighten this scene, punch up this dialogue, proofread the page. It shows you every change it wants to make before any of it counts.

Open it in the screenplay editor, on a Library page, or in any rich-text field:

  • ⌘J / Ctrl+J.
  • The AI Command (wand) button in the editor toolbar, next to the ✨ AI Suggestion button.
  • The wand button in the toolbar that appears over a selection. This is usually the handiest, since most commands act on what you highlighted.
  • Edit → AI Command…

They all open the same Ask AI to edit dialog. If you see ✨ but the wand is dimmed, the two use different models: a suggestion uses the Suggestions model and a command uses the Analysis model, so check both under Settings → AI → Which model does what.

Asking for a change

The dialog has two halves, and you can use either or both:

  • A command. Pick one under Commands, or type to search them.
  • Your own words. "Describe the change — or pick a command below."

Picking a command and typing means both. Tighten plus "and lose the second beat" does exactly that. Press Send.

Type @ in the box to bring in a script, a Library page or a piece of media for the AI to read along with your request.

CommandWhat it does
RewriteReads better, same meaning, voice and format.
TightenCuts every word that isn't doing work.
ExpandDevelops it further in the same voice.
Punch up the dialogueSharpens the dialogue and leaves action and headings alone. Screenplay only.
Fix the formattingElement types, capitals and spacing. No wording changes.
ProofreadSpelling, grammar and punctuation, over the whole document.
Continue from hereWrites on from the cursor.
Find library pagesFinds what the passage names that should have its own Library page. See below.
Generate breakdown reportCloses the dialog and opens the breakdown report.

The last two aren't offered in a form field.

What it is aimed at

Highlighted text is what gets replaced, exactly that. A cursor with nothing highlighted is a place to write. Two commands override this: Proofread always acts on the whole document, and Continue from here always writes at the cursor, even when something is highlighted.

You never have to guess. The dialog says what will happen before you press Send ("Replaces the highlighted text", "Inserts at the cursor", "Applies to the whole document"), and while the command runs, the target is marked in the document too: a filled region where text will be replaced, or a bar on the line where text will be inserted. If a collaborator moves the passage while the AI works, the mark moves with it.

While it works

The document is locked while the AI writes: the cursor stops blinking and typing does nothing. The answer is compared against the exact passage you asked about, and text moving underneath it would spoil that.

To get out, press Escape, press Stop, or use ⌘. / Ctrl+. (the global AI stop). Stopping throws the answer away and leaves your document exactly as it was.

Reviewing the changes

When it finishes, nothing has been committed yet. The changes appear in place in the document, under Review AI changes:

  • Text the AI wants to add is underlined.
  • Text it wants to remove is struck through.
  • A whole new or deleted block gets a bar down its inside edge.

Both readings are on the page at once, so you can see what would be lost as well as what would be gained. The review bar says how many changes there are ("2 of 7") and lets you walk them:

ActionButtonMacWindows / Linux
Next / previous change‹ ›⌘. / ⌘,Ctrl+. / Ctrl+,
Accept this oneAccept⌘ReturnCtrl+Enter
Reject this oneReject⌘DeleteCtrl+Backspace
Accept every changeAccept all⌘⇧ReturnCtrl+Shift+Enter
Reject every changeReject all⌘⇧DeleteCtrl+Shift+Backspace

Changes are split the way a person would split them, by word and by block, not by character, so "Accept" always means something you can read.

You don't have to finish in one go. The document is editable again as soon as the AI is done, and you can keep writing with changes still pending. Anything you type stays yours, whichever way you resolve the change around it.

Undo works normally

Every step is one ⌘Z / Ctrl+Z:

  • The whole set of proposed changes appearing is one step. Undo it and the document is exactly as it was.
  • Each Accept and each Reject is its own step.
  • Accept all and Reject all are one step each.

Nothing the AI does is folded into the sentence you typed just before it, so undo never takes back more than you meant.

Rewriting a speech, a scene, or an action

The selection toolbar offers three shortcuts that skip the dialog: Rewrite this dialogue (with the cursor in a character cue, a parenthetical or a speech), Rewrite this scene (in a scene heading), and Improve this action (in action or a shot).

These act on the whole unit, the entire speech from cue to last line, the entire scene, the whole run of action, whatever you have highlighted. The result arrives as the same reviewable changes as any other command.

Finding library pages

Find library pages reads the highlighted text, or the whole document when nothing is highlighted, and lists what in it has no Library page yet: characters, locations, props, wardrobe, animals, vehicles, set pieces, effects. Nothing is written until you choose. Untick what you don't want and press Create (the button says how many pages). In a screenplay, each new note is also tagged in the scenes that mention it, so it shows on their breakdown sheets.

If the AI read fails, the list comes from a keyword pass alone, and says so. The chat has the same job as a skill (see AI Chat).

Which model runs a command

Commands use the model set for Analysis under Settings → AI → Which model does what, not the one for Suggestions. A suggestion is something you sit and wait for, so that row prefers the fastest model; a command is a considered rewrite and is better served by a stronger one.