Purpose
Hotel Lobby AI is used to turn two portrait photos into the popular "Hotel Lobby" two-person duet video (Hotel Lobby AI duet). The scene is set in an orange recording booth, with two performers standing on either side of a hanging microphone, one singing first, the other responding, and then switching. The official description says that friends, partners, or family members can appear in the same video even if their original photos were taken separately.
Main capabilities
- Upload two photos, designate the performers appearing on the left and right respectively, and optionally add a one-sentence styling note (such as clothing, accessories, or expression).
- Use GPT Image 2 to composite the orange recording-booth scene, then use DreamActor motion transfer to transfer the reference performance and its audio onto the generated people.
- Output a vertical 9:16 MP4 video with audio, with a choice of classic shot or full-body shot.
- The official claim is that results take about 3–5 minutes; the page mentions that after clicking generate, the scene is ready in under a minute, and the finished video is completed in about 3–5 minutes.
- Finished videos are saved in "My videos."
How to use
The officially given process is:
- Choose a clear portrait photo for each side (visible face, natural expression, and similar camera angles help keep the two people distinguishable).
- Confirm consent and begin generation.
- You can write a simple sequence of events: A starts first, B responds, then they swap who leads; a fixed camera and smaller movements usually make it easier to judge the first generation result.
- Click generate and choose a package; the system first composites the two people into the orange recording booth, then transfers the Hotel Lobby performance and audio.
- Before sharing, check identity consistency, hands, the central microphone, and left/right positions.
The page also provides a copyable starting prompt idea: vertical 9:16 framing, matte orange recording booth, two separate people, one black microphone, soft lighting, and a fixed camera, and requires the model to preserve faces and clothing while avoiding left/right swapping, adding extra people, subtitles, and watermarks. Styling details are written in the direction field, kept short to avoid affecting the core composition.
Use cases
The officially given usage suggestions include:
- A duet between friends: one person sings the first line, the other responds at the microphone, using eye contact, smiles, and small gestures to show the relationship.
- A back-and-forth between siblings: one starts confidently, the other joins on the next beat, standing on either side of the orange recording booth.
- A couple'sSynchronized performance: give a quiet first section and a shared chorus, keep movements small, and mark the moment of convergence with eye contact across the microphone.
- It can also be used for couples, friends, or family (including parents and their children, provided the uploader is the child's parent or legal guardian).
Fees and credits
The official page states "pay-per-generation · no subscription required," and credits are consumed only when generating duet videos:
- 8-second duet video = 440 credits
- 15-second duet video = 680 credits The credit tiers listed on the page roughly correspond to: one tier is about 1 8-second duet video; one tier is about 2 8-second videos or 1 15-second video; one tier is about 5 8-second videos or 3 15-second videos; one tier is about 10 8-second videos or 6 15-second videos. You can choose a one-time credit pack or a monthly subscription. If generation fails, credits are automatically returned to your balance.
Background and limitations
The official description says this trend originated from the video of Quavo and Takeoff performing "Hotel Lobby," released on the COLORS YouTube channel on June 17, 2022: the two stood in front of a solid orange background, sharing a single microphone suspended from the ceiling and taking turns singing. Starting in September 2026, creators on TikTok and X began using AI video tools to replace the two rappers with friends, couples, celebrities, fictional characters, or even pets, while preserving the original movements, lip sync, and audio.
Limitations and notes:
- Identity preservation depends on the generation model and the photos, so results will vary; the model may drift when references are unclear or movements are complex, and multiple attempts may be needed to get a usable clip.
- The official guidance recommends using photos with clear, complete faces and recognizable clothing, and avoiding heavy filters, sunglasses, crowded backgrounds, and images with the face cropped out; two reference photos with similar lighting and angles produce more consistent results.
- Only images authorized for use may be transformed, and permission must be obtained from every person in the photos; when sharing, label the content as AI-generated.
- The two uploaded photos are resized in the browser and, after clicking generate, are sent to the AI provider fal.ai to create the scene and video; they are not uploaded before clicking generate.
- The official statement says all generations consume credits, with the specific amount shown on the button and in the pricing section; the page also recommends checking the video before sharing, including confirming that A stays on the left, B stays on the right, and the microphone is between the two people, checking for hand deformities, merged faces, extra people, or clothing changes, and watching once with the sound off to confirm that the handoff still holds up visually.
