Hotel Lobby AI prompt
Short answer: with a template tool like hotellobby.dev you do not need a prompt. If you use a general image-to-video tool, adapt the text below.
Adaptable prompt
This is the structure we use. It works best when you write the exact lines yourself, because a prompt that only says "they rap" makes most video models invent unintelligible speech.
Two performers standing side by side in a small minimal orange recording booth, one hanging microphone between them, fixed camera, vertical 9:16 frame with both performers fully visible, warm studio light. The performer from Image 1 is on the left and the performer from Image 2 is on the right. They perform in two alternating turns of about five seconds each. In the first turn the performer on the left looks at the camera and says exactly: "[your first line]" while the performer on the right listens with a closed mouth. In the second turn the performer on the right says exactly: "[your second line]" while the performer on the left listens with a closed mouth. Each performer has a distinct voice that matches their appearance. Between the lines both performers are silent. All words are clear English, exactly the lines given, no extra words and no made-up language. Keep each identity exactly as in the reference image. No cuts, no camera movement.
Why it is written this way
- Write the lines. Without lines, models fill the audio with babble that sounds like speech but is not.
- Take turns. One performer speaks while the other listens. Two people speaking at once blurs both faces and both voices.
- Keep lines short. About 10 to 15 words per turn fits five seconds. Longer lines get rushed and slurred.
- Ask for silence between turns. Otherwise the model adds ad-libs in the gap.
- Use original words. Do not paste lyrics from a real song or imitate a real performer.
Voice styles that work
The delivery changes the whole clip. Styles we have tested with scripted lines: a fast rap exchange, a slow ballad, deadpan one-liners, a soft whisper, a tired sarcastic grumble, and two pets hosting a podcast. In the generator you pick one and the lines are already written, so you can skip the prompt entirely.
Tips
- Use two reference images, one per performer, and describe each briefly.
- Keep the camera fixed; moving cameras break identity.
- Short clips (about 10 seconds) hold faces better than long ones.
- Use adults, pets or fictional characters only, never children or celebrities.
General tools need several tries to get a stable result. A template removes that trial and error: use the generator.