Write Positions the World Can Read

Quick answer: Positions come from the storyboard text. Write them as relations to things in the room and to each other: at the head of the table, across from her, in the doorway, with his back to the window. Say the posture when it is not standing, and who faces whom. Write it once at the start of the segment, and again only when someone moves. Prefer room anchors over screen directions: “by the window” holds in every shot, while “on the left” is only true from one camera. Where you say nothing, people are still given a sensible place and kept there.
Still from “Rust City Knockout”, an AI short drama generated end-to-end by SceneMixer (6 characters, 1 episode)
Sample: “Rust City Knockout” — Painterly Anime, 1 episode, produced end-to-end by SceneMixer from the script · watch the episodes

Open a segment in the storyboard and read its opening shot. If it does not say who is where, add one sentence, such as @Mia sits at the round table by the window; @Leo sits across from her, his back to the bar.

Where positions come from

When a segment is generated, every character in it is given a place in the scene’s room, read from what the storyboard says. On projects with the world model that room is the measured set built from the scene image; on other projects it is the earlier automatic staging. Either way, the text is the input.

Everyone in the scene always has a place. If the text says nothing about someone, they are put somewhere sensible and kept there, but “sensible” may not be what you pictured. One sentence settles it.

Anchor people to the room, not to the frame

“On the left” only means something from one camera. Turn the camera around for the reverse shot and the left is now the right. “By the window” means the same thing from every camera, because the window is part of the room.

A phrasebook for common places

PlacePhrases that place people clearly
Tablesits at the head of the table · sits across from @Leo · sits beside her, on the window side · stands behind his chair
Bar or countersits at the end of the bar · leans on the counter facing the room · stands behind the bar
Doorstands in the doorway · waits just inside the door · leans against the closed door
Windowstands at the window, looking out · sits with her back to the window
Sofa, bedsits on the sofa next to @Mia · sits on the edge of the bed · lies on the bed, facing the wall
Desksits behind the desk · stands in front of the desk, facing him
Stairsstands at the top of the stairs · sits on the bottom step
Carin the driver’s seat · in the passenger seat · in the back seat behind the driver

Cars, buses, trains and planes have their own guide: Scenes inside a car, bus, train or plane.

Posture: say it when it is not standing

Standing is assumed. Anything else should be in the text, in the shot where it starts: sitting, kneeling, crouching, sitting on the floor, lying down, seated in a vehicle. A figure that should be kneeling but is written as nothing will stand.

While a segment renders, the 360° preview draws each person in the pose that was read: Look around the set while a shot renders.

Facing and eyelines

Say who faces whom, and who has their back to what. That decides which way each person looks, and so which cameras can see their face.

Why this matters at the cut: Why characters swap sides between cuts.

Movement: start, destination, one move per shot

A move needs a destination the room has. “He moves” says nothing; “he walks to the door and stops in the doorway” gives a place to end in, and that place holds for the next shots.

Outdoors, crowds and checking your work

Outdoor scenes get a lighter world: the ground, the nearby objects and the distance in four directions. Anchor people to objects that are really there, such as the bench, the fountain or the car, rather than to open ground. Background people are written separately from the named cast: Background people in a scene.

To check: on projects with the world model, select a segment that has not been generated and drag around the set to see what the room holds. While it renders, the figures show where everyone was placed. If someone is wrong, fix the sentence and generate again once the run finishes.

Instead ofWrite
They talk.@Mia sits across from @Leo at the round table.
On the left of the screen.By the window. / With her back to the bar.
He moves.He walks to the door and stops in the doorway.
She is down.She kneels beside the bed.
In the car.In the passenger seat, beside @Leo who drives.
Somewhere in the room.At the end of the bar, facing the room.

FAQ

Do I need to write positions in every shot?

No. Write them once at the start of the segment, then again only when someone moves or changes posture.

What happens if I write nothing about positions?

Everyone is still given a sensible place and kept there across the shots. It may just not be the place you pictured.

Can I use left and right?

You can, but they change with the camera. Anchors in the room, like the window or the door, mean the same thing from every angle.

What if the scene image does not have the thing I mention?

Then there is nothing in the room to put the person by. Anchor them to something the image shows, or make a new scene image that has it.

Why is someone standing when I wrote that they sit?

Check that the posture is in the text of the shot where it starts, and that the person is named with an @ chip there. Then regenerate the shot.

Does writing positions cost anything?

No. Positions are part of the storyboard text; you pay only for what you generate.

How many people can I place in one scene?

Everyone named in the segment gets a place. For a crowd behind the main characters, describe background people instead of naming each one.

Can I drag people into place instead of writing it?

No. The 360° preview is read-only. The text is the way to place people.

One sentence puts everyone in place

Open your storyboard, add who sits where and who faces whom, and generate.

Open my projects

Updated October 15, 2026

Pricing·Novel to Video·Guides·Samples·Alternatives·Answers·Support·Privacy·Terms·Legal·Contact·Affiliate program·© 2026 SceneMixer