Hotel Lobby AI Prompt: Keep Two Faces and Roles Stable

A good Hotel Lobby AI prompt does four jobs in order: map the two photos, lock the camera, describe one exchange, and direct original sound. Put identity first. If the first paragraph is all lighting and atmosphere, the model has already started composing before it learns who belongs where.

Use this as a starting prompt for a vertical ten-second clip:

Image 1 is the left performer. Image 2 is the right performer.
Keep both people visible in a fixed vertical medium-wide two-shot.
Preserve each person's face, hairstyle, skin tone and clothing from
their own reference photo. Do not merge, replace or swap identities.

They stand in an original amber-orange recording booth with one plain
studio microphone hanging between them. Image 1 nods to the beat and
delivers one short original line. After a brief pause, Image 2 answers
with a different short original line. They finish with one natural nod
to the camera. Use small hand gestures and no fast choreography.

Audio: an original minimal hip-hop beat, the two original vocal phrases,
and quiet booth room tone. No existing song, recognizable lyrics, artist
voice, celebrity likeness, logos, captions or on-screen text.

The Hotel Lobby AI Generator accepts two portrait references and an editable prompt like this one. It generates a new scene rather than replacing faces in a source music video.

Why the first two lines matter most

“Image 1 is the left performer” is stronger than “put my friends in a booth.” It ties the upload order to a visible position. The next line keeps both faces on screen, which gives the model less reason to cut to a solo close-up and forget the other reference.

Keep role words consistent. Do not call Image 1 “the left performer,” “the taller friend,” “the person in red,” and a name in four different sentences. Pick one label and repeat only when the action changes hands.

Fix swapped or blended faces

If the faces trade places midway through the clip:

1. move the mapping sentence to the very beginning;
2. remove camera rotation and crossing movement;
3. keep each performer on their original side;
4. ask for a fixed medium-wide two-shot;
5. shorten the action before changing the resolution.

If the two identities start to blend, check the photos as well as the words. Separate front-facing portraits with different backgrounds are easier to map than two faces cropped from the same group picture. Avoid references where one face is dramatically smaller or partly hidden.

Do not try to fix identity by repeating a person’s name ten times. The model has the pixels, not a verified identity record. Spatial labels and a simpler shot give it more useful information.

Fix awkward hands and gestures

Finger-counting, pointing directly at the lens, crossing arms in front of the face, and complicated handshakes all create brief occlusions. Start with a shoulder nod, an open hand, or a single fist bump. These gestures still read on a phone screen and leave the faces visible.

For example, change:

They perform an elaborate synchronized handshake while the camera orbits them.

to:

The camera stays fixed. After Image 2 finishes the answer, they make one
simple fist bump at chest height, then lower their hands and face the camera.

That version tells the model when the gesture happens, where it happens, and when it ends.

Keep the vocal turn short enough to hear

Two full verses do not fit inside ten seconds. Write the intention rather than packing a paragraph of lyrics into the prompt. One compact original line per person is enough to establish the call-and-response.

If the words overlap, add sequence language:

Image 1 speaks first. Pause for one beat. Only then does Image 2 answer.

If you care more about visual identity than exact speech, direct vocal energy instead of quoted dialogue: “Image 1 delivers a confident short phrase; Image 2 answers playfully.” You can finish the writing and mix in an editor after the visual take works.

The audio request should name sound sources that can exist in the scene: beat, voice, room tone, mic handling, or an elevator chime. “Make it cinematic” says almost nothing about what the listener should hear.

Adapt the prompt without turning it generic

The visual setup is recognizable because it is spare. Personalize the reason for the performance, not just the color:

• two friends trade a birthday line;
• coworkers celebrate shipping a project;
• siblings joke about who tells the same family story;
• a couple announces a trip;
• teammates mark a win.

Each idea gives the clip a shareable purpose. It also produces more original writing than asking for an artist impersonation.

Five seconds is enough for one shared reaction. Ten seconds is the best starting point for two short vocal turns. Use fifteen only when you can name the third visible beat. More time without more structure usually creates an ending that wanders.

When the result is close, change one variable at a time. Swap the photo, simplify the gesture, or tighten the vocal sequence—do not rewrite the cast, camera, action, sound, and duration together. A prompt becomes useful when it lets you diagnose the next render, not merely describe the mood of the last one.