How to Design a Two-Character AI Video Scene

01 / CAST DESIGN

Give each fictional adult a distinct visual identity

Two characters are easier to distinguish when their descriptions do not compete for the same cues. Start with two clearly adult, fictional people and choose differences a viewer can see at a glance: silhouette, hairstyle, clothing palette, posture, and a small personal prop.

• Mara, 34: tall, close-cropped dark curls, a rust-colored overshirt, and a folded notebook.
• Ivo, 38: shorter, wavy silver hair, a deep-blue jacket, and a brass key ring.
• Keep their faces, clothes, and props distinct in every written description. Do not base the characters on real people without permission.

These are original fictional designs for this exercise, not portraits, customer examples, or generated outputs. Use a reference image only if the service you choose supports it and you have rights to use it; this guide does not assume multiple-image input.

02 / SCENE DESIGN

Block the interaction before adding camera direction

Decide where each person stands, what each one wants, and which person handles the important object. Give them different actions so the scene has readable roles instead of two descriptions layered on top of one another.

1. Mara stands on the left at the front edge of a desk in a small community radio studio, holding a folded notebook.
2. Ivo stands beside her on the right, within arm’s reach, with the brass key ring in one hand. The open studio door remains behind him.
3. Ivo passes one key to Mara; she takes it and looks down at the notebook.
4. Both remain in the same positions and clothes while the camera holds a medium two-shot.

A single interaction is a useful starting point. Add camera movement only after the blocking is clear, and describe one view or move at a time. Avoid simultaneous actions that make it unclear who touched or moved the prop.

03 / PROMPT EXAMPLE

Write one prompt with explicit roles and positions

Original prompt: “One continuous medium two-shot inside a quiet community radio studio at dusk. Two fictional adult characters are present: Mara, 34, tall with close-cropped dark curls, wears a rust-colored overshirt and holds a folded notebook; Ivo, 38, shorter with wavy silver hair, wears a deep-blue jacket and holds a brass key ring. They stand side by side at the front edge of the desk, within arm’s reach: Mara on the left and Ivo on the right, with the open studio door behind him. Ivo removes one key from the ring and extends it toward Mara. Mara takes the key, then glances at her notebook. Keep both characters in their starting places and keep the exchange visible between them. Preserve each person’s distinct appearance, clothing, and props. Keep the room and warm window light steady. The camera stays at eye level and does not cross the space between them.”

This text describes a scene; it does not assert that a particular model can preserve both identities, accept several references, or obey every blocking instruction. Check the actual service’s supported inputs and evaluate the returned clip.

04 / ITERATION

Review identity, blocking, and the handoff

Watch the full clip and check each character separately. Confirm who begins on each side, which person holds the key, whether the handoff is visible, and whether the clothing cues remain distinct. A still frame can look correct while the action becomes ambiguous in motion.

• If the faces or outfits converge, strengthen the most useful visual contrast and remove overlapping wardrobe details.
• If the key changes hands too early, describe the starting holder and the single handoff more plainly.
• If the characters swap sides, simplify movement and restate their starting positions.
• If the camera move hides the exchange, return to a stable two-shot before adding a new camera direction.

Change one cause at a time and keep a note of what the next attempt is meant to test. If the tool does not support the needed scene or reference controls, choose a simpler shot or another verified workflow rather than implying that a prompt creates missing features.