Prompting Two People in a Hotel Lobby AI Video

Two-person clips fail more often than solo clips, and most failures are identity problems: faces blending, people swapping sides, or both subjects moving in sync. This post covers the parts of a prompt that matter only when there are two people. It assumes you already know the basic structure from the prompt guide.

Assign sides explicitly

Name a fixed side for each reference image: "reference image 1 is the person on the left, reference image 2 is the person on the right." Do not rely on the model to decide. Ask for both people to stay on their sides for the whole clip, so they do not swap places when they trade the lead.

Give each person a different job

The original COLORS performance is a back-and-forth. One person leans toward the microphone while the other bobs, points or adds ad-libs. LightX warns that when both subjects do the same moves at once, the video stops feeling like a duo. Write the first half and second half separately:

First half: left person leads, right person nods and points.
Second half: right person leads, left person steps back and bobs.
Ending: both lean toward the microphone together and hold a small smile.

Make the two people look different

If your two reference photos have similar hair, clothing or lighting, the model is more likely to blend them. Choose photos where the two people differ in at least one obvious way, such as hair, clothing color or height. LightX also advises separate photos, one per person, rather than one group photo.

Keep the frame simple

One microphone, hanging between them.
A solid orange wall with no props or text.
A locked camera. Movement makes faces drift.
Waist-up or full-body framing, but not a mix.

Phrases that help

"The two people stay distinct for the whole clip."
"Each person keeps their own face, hair and clothing."
"No merged faces, no swapped sides."

Phrases that hurt

Avoid asking for both people to "rap together in perfect sync," since that creates the same-motion problem. Avoid long lists of camera moves. Avoid asking the model to render the song's lyrics. See copyright and safety.

More examples for pets, solo and character versions are in the prompt examples post.

Practical points

Use this topic as part of the wider Hotel Lobby AI workflow: identify the original reference, decide what changes, generate a short test, inspect faces and motion, then check consent, labeling and platform rules.

Separate the 2022 source performance from newer AI-generated clips.
Use clear reference images and consistent framing.
Inspect hands, faces, lip movement and identity drift.
Use current platform disclosure controls where applicable.
Do not use private people's images without appropriate permission.

Follow Us