我们注意到您可能偏好使用 简体中文. 要切换语言吗?
Il semble que vous préfériez Français. Souhaitez-vous changer de langue ?
Es scheint, Sie bevorzugen Deutsch. Möchten Sie die Sprache wechseln?
Parece que prefieres Español. ¿Te gustaría cambiar de idioma?
Sembra che tu preferisca Italiano. Vuoi cambiare lingua?
優先言語は次のようです 日本語. 言語を切り替えますか?
선호하시는 언어는 다음과 같습니다 한국어. 언어를 변경하시겠습니까?
Похоже, вы предпочитаете Русский. Хотите сменить язык?
Tercih ettiğiniz dil Türkçe. Dili değiştirmek ister misiniz?
Wygląda na to, że wolisz Polski. Czy chcesz zmienić język?

Black Friday: Up to 66% off most AI models.

Explore the deals
← Blog

Guides

Hotel Lobby AI Video: References, Prompts, and a Two-Person Workflow

Plan a two-person Hotel Lobby-style video with identity and motion references, an illustrative prompt, and checks for faces, gestures, and audio.

APIXO Team

Start creating
Two generic performers on an orange stage with identity, motion, and prompt reference roles

To plan a Hotel Lobby AI video, give each source a job: two image references define the performers, a motion reference guides the performance, and your prompt describes the scene and what should stay consistent. Review the video before adding or approving the final audio.

This guide uses original editorial diagrams and external examples. The prompt is illustrative and has not been tested in a paid APIXO generation; the embedded remixes are other creators’ work.

What is the Hotel Lobby AI video trend?

The format borrows the orange set, shared hanging microphone, and two-person performance from Quavo and Takeoff’s “HOTEL LOBBY” on A COLORS SHOW. The song gives the trend its name. Start with the official COLORS performance to understand the framing.

official COLORS performance

https://www.youtube.com/watch?v=x9yop0nYR9g

Original performance by Quavo & Takeoff, shared by Quavo on X. Observe the performers’ positions, shared microphone, and camera framing. This is the source performance, not an AI result. View the original post on X · Full performance on YouTube

View the original post on X

https://x.com/QuavoStuntin/status/2102798534173233453

Changing the cast creates the remix. When watching a version, separate what you can observe in the output from the creator’s production method. A polished clip alone does not reveal which model, prompt, or number of attempts produced it.

External remix by @ogfeitoza. Compare the two character positions and the recognizable orange-stage composition. This post is an example of the format, not evidence of APIXO output quality. View the remix on X

View the remix on X

https://x.com/ogfeitoza/status/2100991271695192335

Choose the result you actually want

  • Goal: An original orange-stage duet · Useful inputs: A composed starting image or character references, plus a written action brief · What to evaluate: Recognizable subjects, a readable duet, and coherent movement
  • Goal: A remix following a particular performance · Useful inputs: Two identity references and a usable motion-reference clip · What to evaluate: Which person follows which role, gesture timing, and camera framing

A prompt can describe the scene and intended action. When the exact sequence of gestures matters, provide motion information instead of asking prose to carry the entire choreography. A reference still guides generation; it does not guarantee a frame-by-frame match.

1. Prepare the identity and motion references

  • Choose one clear image for each performer. Favor visible faces and readable clothing over dramatic lighting or heavy occlusion.
  • Write down who stays on the left and who stays on the right. Use the same assignment in your reference notes and prompt.
  • If the performance includes turns, consider consistent additional views of each character. Check that the face and outfit remain the same across those views.
  • Choose a motion clip in which both roles are easy to follow. For an original performance, you can film your own short duet with a steady camera. Each reference video must be 2–30 seconds, and the total reference-video duration must not exceed 30 seconds. Trim a longer source to a usable segment before uploading.
  • Decide where the final audio will come from before generation: an original or licensed track, generated sound, or a later edit.

Use material you can use for this project. Watching or embedding the official performance does not make its footage or music an automatically licensed generation input. The embeds in this article are viewing references.

Original editorial diagram: assign a role to each input, then review and revise. These are planned stages, not a recorded generation history.
Original editorial diagram: assign a role to each input, then review and revise. These are planned stages, not a recorded generation history.

Original editorial diagram: assign a role to each input, then review and revise. These are planned stages, not a recorded generation history.

2. Set up the workflow in APIXO

Open Seedance 2.5 on APIXO. In the Playground, set Mode to Omni Reference when you need character images and a motion clip together. This reveals Reference Images, Reference Videos, and Reference Audio. The API documentation calls this mode omni_reference. Sign in if the page requests it.

  • Assign the image references to performer A and performer B, and the video reference to motion. Make the asset-to-role mapping explicit in your prompt.
  • Choose the output framing before generating. For a vertical post, confirm that both people, their hands, and the microphone will remain in frame.
  • Use a defined shot length for an initial test. If you use dynamic duration, check the documented pre-charge and settlement rules before estimating your balance.
  • Decide whether to enable Generate Sound. Generated sound and the original song are different inputs; do not assume enabling sound reproduces the source recording.
  • Review the current estimate and input limits before pressing Generate. Record the prompt and settings so a later revision has a clear starting point. With video references, billable seconds include both the reference videos and the output video.

The public controls include Mode, Prompt, Draft, Resolution, Duration, Aspect Ratio, and Generate Sound. Available reference controls depend on the selected mode. The instructions here describe setup; they do not report a successful Hotel Lobby run.

3. Write the prompt around roles and motion

Treat this example as a brief to adapt to your assets. A and B are labels in the explanation, not special API fields. Match them to the actual references you supplied.

Shared English prompt example

Create one continuous two-person performance using the supplied references.
Identity reference A defines the performer on the LEFT.
Identity reference B defines the performer on the RIGHT.
Use the motion reference to guide each position's gestures, body rhythm,
head turns, timing, and camera framing.
Preserve each reference person's facial features, hairstyle, and outfit.
Keep the two identities separate throughout the shot.
Set the performance in a plain burnt-orange studio with one silver
microphone hanging between the performers.
Keep both people visible. Keep the camera steady and the shot continuous.
Use natural body movement and consistent studio lighting.
No extra performers, no captions, no logos, and no scene cuts.

The first lines establish identity and position. The next lines assign movement to the video reference. The final lines define the environment, camera, and exclusions. Describe a coherent performance first; do not add several unrelated camera moves or scene changes to the same short shot.

Original planning storyboard: establish the duo, alternate the lead, and finish with both visible. It illustrates an original performance structure, not frames from a generated clip.
Original planning storyboard: establish the duo, alternate the lead, and finish with both visible. It illustrates an original performance structure, not frames from a generated clip.

Original planning storyboard: establish the duo, alternate the lead, and finish with both visible. It illustrates an original performance structure, not frames from a generated clip.

4. Review the result and choose one revision

APIXO’s Seedance 2.5 model notes say references do not guarantee exact identity or motion preservation. Watch the complete clip once, then inspect the moments where the brief places the most demands on the model.

  • If you notice…: Faces merge or the performers swap roles · Check first: Image clarity, role assignment, and contradictory identity instructions · A next test to consider: Clarify left/right roles and remove conflicting references
  • If you notice…: The scene looks right but gestures differ · Check first: Whether you supplied a motion reference and whether both roles are readable · A next test to consider: Try a shorter, clearer motion segment
  • If you notice…: Hands distort when crossing faces · Check first: Occlusion, gesture speed, and how much detail the shot must retain · A next test to consider: Reduce gesture complexity in an original performance brief
  • If you notice…: The duet loses a performer in a vertical crop · Check first: Composition of the inputs and the requested output framing · A next test to consider: Prepare a reference with room for both subjects
  • If you notice…: Mouth movement or gestures miss the beat · Check first: Video timing, generated sound, and the separately added audio track · A next test to consider: Check the edit’s start point and compare the entire sequence

These are checks and possible next tests, not validated fixes. Save the previous prompt and output, change one major instruction or reference, and compare against the same acceptance criteria. If the problem persists, reconsider the motion or composition rather than assuming more prompt text will solve it.

5. Finish the audio and delivery version

After adding the reference track, review the complete new version with sound. If you choose a different track or continue editing elsewhere, manually check its starting point, head turns, gestures and mouth movements. A matching first frame does not establish synchronization throughout the clip.

For multiple directions, see Seedance Draft-to-Final. Drafts and Finals each use Credits. Use the video budget guide to plan revisions, then check the Credits estimate for your selected settings before generating.

Common questions

Can I make this with only two photos?

Two photos can communicate who appears. For an original duet, combine them with an action brief or prepare a composed scene. If you want a specific performance’s gestures and timing, a motion reference gives the model information the photos do not contain.

Does the prompt include the original Hotel Lobby song?

No. This example describes the visual performance. Choose your audio source separately and confirm its suitability for the intended use.

What if reference music triggers a copyright check?

If a Seedance task fails a copyright check with sound enabled, check the music in your reference audio or video first. Protected reference music, or generated music that closely resembles an existing song, can trigger the check.

  1. Use the failed card's option to generate a silent version. It keeps the prompt, reference images, audio and videos, and turns off sound output for a new generation. This retry is charged normally; review the estimate first. The option appears only for tasks that had sound enabled and failed a copyright check. Turning sound off does not guarantee success.
  2. Once the silent video succeeds, use the result's option to add the reference audio track. APIXO uses your uploaded audio first, or the reference video's original sound if you uploaded no audio. This step is free and combines the track with the video without regenerating the visuals. The silent version remains, and a new version with sound is added.

There must be one clear audio source: exactly one reference audio file, or no audio files and exactly one reference video. Multiple audio files, or multiple videos without an audio file, make this option unavailable. Without either type of reference, there is no reference track to add.

Turning off sound output does not remove the reference audio's timing information. The model can still use that rhythm to guide motion, so adding the same track usually avoids aligning it from scratch. Play the full result to check gestures, beats and mouth movements; rhythmic alignment does not guarantee word-perfect lip sync.

If the failure concerns visual content safety, such as a celebrity likeness, change the reference images or video, or revise the prompt. Turning sound off cannot resolve a block caused by the visuals.

Where can I learn a more detailed reference workflow?

Kapwing’s Hotel Lobby YouTube tutorial (English) explains its character-sheet, simplified motion-reference, and editing approach. Those are Kapwing’s steps; APIXO’s available inputs and controls should be checked on its own page. For basic movement prompting, see APIXO’s motion-prompt guide.

Kapwing’s Hotel Lobby YouTube tutorial (English)

https://www.youtube.com/watch?v=MWurz-UUCH0

Start with a clear duet brief

Choose the two identities, define the left and right roles, and decide how closely the action needs to follow a reference. Open Seedance 2.5 with that brief, review the estimate, and evaluate the output against identity, motion, framing, and sound.

Sources and reference examples

Product capabilities and public controls checked October 9, 2026. External examples do not represent APIXO tests.

Build with APIXO

Explore image, video, and audio models in one place.