How to direct AI actors with text

Updated September 13, 2026 · 6 min read

Directing an AI actor is writing a stage note, not a prompt: a good direction line names who, where, what changes, and the one held beat — then lets the Director stage it. Bad lines describe moods ("make it dramatic"); good lines describe actions ("she reads the letter twice before looking up"). This guide breaks the line into five parts and shows the fixes for the five most common misses. First real direction takes minutes.

Time to first take: ~5 minutes
Difficulty: Beginner craft
You need: A Cast + a World
Studio surface: Director

The Director panel with a typed direction line — who, where, what changes, the held beat — staged into a scene. — One typed line is a direction. This guide teaches the five parts.

Learn the five parts of a direction line
Every line that stages cleanly has the same five parts: who (name the Cast member, don’t describe them), where (the set or position), what happens (an action, not a mood), what changes (the turn inside the scene), and the held beat (the image to end on).
“Mei Lin at the terrace door, reads the letter, reads it again, then looks up at Jun-ho — hold on her face” is a complete direction. It’s stage directions, not prose — the Director turns it into blocking, shots, and timing.
Who, where, action, turn, held beat — the five parts.
TRY THIS
Describe your show
Fix the mood-word problem
The most common miss is the mood word: “make it tense,” “keep it dramatic.” The Director can’t stage an adjective — it stages verbs. Tension is a verb wearing a costume: she doesn’t sit down, he keeps the letter folded, nobody says the name.
The fix is mechanical: take your mood word and ask what a camera would see. “Tense” becomes “she stands while he sits; she doesn’t touch the tea he poured.” If you can’t see it, neither can the Director.
Verbs, not adjectives — the camera can’t see “tense.”
Name the held beat — the shot the scene ends on
Scenes don’t stop, they land — and you choose where. The held beat is the one image the Director holds past the action: her face after the news, his hand not reaching for the phone, the door nobody walks through.
Write it into the line: “end on the letter in her hand.” If you don’t name it, the Director picks the last action instead of the last image — and actions resolve, while images withhold. The cliffhanger guide is the deep version of this one skill.
The held beat: the frame that outlasts the scene.
Direct one change per retake
A retake isn’t a new scene — it’s the same scene with one note. When a take misses, change exactly one thing: the position, the timing of the turn, or the held beat. “Same scene, but she’s already seated when he enters” is a note; retyping the whole scene is a re-roll.
One-change retakes keep the parts that worked — the blocking you liked survives, and you can actually tell which fix landed. Two or three noted passes is a scene; eight blind regenerations is a slot machine.
Same scene, one note per take — the blocking survives.
Keep what lands — the cut assembles itself
Directing ends in keeping, not exporting. Every take you keep drops into the episode in order; the app cuts your kept Clips together with no timeline to learn. Your job is selection: the take where the held beat lands is the take that survives.
And screen for keeps, not for curiosity — every completed take is charged whether you keep it or not, and the price is on the button before you press. A director who screens honestly finishes a scene in three takes, not nine.
Kept takes assemble themselves into the episode.
TRY THIS
Describe your show

Made this way

One line staged: two people, one object, one refusal.
Blocking you can read: position is a direction too.
Same grammar in anime: who, where, the turn, the hold.
A comedic held beat — same five parts, faster tempo.

Generated in Character App · example stills, not a live product demo

Common mistakes

Adjectives instead of verbs. “Make it dramatic” gives the Director nothing to stage — describe what a camera sees, not what a viewer should feel.
Re-describing the characters in the line. The Cast card already knows who they are — name them, never re-roll them.
No held beat named. Without “end on X,” the scene ends on the last action instead of the last image — and actions resolve.
Rewriting the whole line on a miss. A retake changes one thing; a rewrite re-rolls everything, including the parts that worked.

Questions

What’s the difference between a direction line and a prompt?

A prompt describes a picture; a direction line describes an event. Prompts produce stills — direction lines name who, what changes, and where the scene lands, so the Director can stage them.

How long should a direction line be?

One sentence, five parts. If it needs two sentences, it’s probably two scenes — the minute an episode guide covers the split.

Can I write dialogue in the direction?

Short lines, yes — “she says the name out loud” stages fine. Long exchanges work better broken across scene beats than embedded in one direction.

Why did my direction come back different than I pictured?

Usually an unstaged mood word or a missing held beat. Check the five parts — whichever one is absent is the one the Director improvised.

How many retakes is normal for one scene?

Two or three noted passes. Each retake is a charged Clip at the price on the button — change one thing per pass and you’ll rarely need more.