# How to make a Hotel Lobby AI video from two photos

By SeedRouter · Published 2026-10-10 · Updated 2026-10-10

To make a **Hotel Lobby AI** video on SeedRouter, keep or replace the prefilled left and right performer images, review the supplied motion video, then generate a 15-second duet. The tool adds the supplied soundtrack after generation and provides a download containing both video and audio.

Open the [Hotel Lobby AI Playground](https://seedrouter.ai/tools/hotel-lobby-ai#playground) to follow these steps. This is a workflow guide, not a claim that every pair of photos will produce the same gestures or likeness. Check the whole result before sharing it.

## What goes into the duet?

The scene prefills two character images, a motion clip and an editable prompt. The three media inputs appear in their corresponding upload boxes. Audio generation is turned off (`generate_audio: false`), and the soundtrack is added only after the silent video is generated. It is not sent as a reference audio input. Each input has a different job:

| Input                    | What it supplies                         | What to check                                                |
| ------------------------ | ---------------------------------------- | ------------------------------------------------------------ |
| Left performer, Image 1  | The first character’s appearance         | One clear face, visible hair and outfit                      |
| Right performer, Image 2 | The second character’s appearance        | The intended character is in the correct slot                |
| Motion reference         | Gestures, framing and performance timing | Preview the included line-art clip                           |
| Prompt                   | Role assignment and the requested scene  | Avoid instructions that conflict with the reference          |
| Soundtrack               | Sound for the finished download          | Added after generation, separately from the reference inputs |

The photos are reference images, not first and last frames. Those are different input modes in Seedance: frame inputs cannot be combined with the image and video references used by this preset. \[1]

## 1. Prepare two character photos

Start with one character per image. A face hidden by sunglasses, a cropped hairstyle or a crowded background gives the model less useful information about the intended performer. Choose images in which the face and clothing are clear, and put each character in the corresponding left or right slot.

Use photos or original characters you have permission to use. If you are trying an animal duet, an image that already shows the intended upright posture can make your instructions clearer. This is a preparation suggestion, not a promise that the output will preserve every detail of a pet’s appearance.

The sample performer images are already loaded. Remove a sample to use your own image. The uploader accepts files or public image URLs. Uploaded files are turned into URLs before the generation request is submitted. You can add supporting reference images after the two main performers; keep their order clear so Image 1 and Image 2 retain their assigned roles.

## 2. Keep the motion reference and review the prompt

The Hotel Lobby AI preset includes a roughly 15-second line-art motion clip. Play it before generating: it shows the turns and gestures you are asking the new characters to follow.

The editable prompt starts by assigning the performers and separating appearance from movement:

> Image 1 is the LEFT performer; Image 2 is the RIGHT performer. Preserve each character’s face, hair, clothing and identity from their own image throughout the shot. Use Video 1 only for choreography, body gestures, mouth movements, camera framing and performance timing.

The rest of the preset asks for a single continuous performance at one microphone against a warm orange backdrop. It asks the model to retain the visual style of the uploaded character images instead of copying the reference’s line-art appearance.

Keep those roles intact for a first attempt. If you change the prompt, make the desired difference specific: an outfit detail or a facial expression is easier to inspect than several competing instructions about camera cuts, new locations and additional characters. **Reset restores the scene preset**, including its prompt, both performer images, reference video.

## 3. Choose the model and inspect the estimate

The tool offers [Seedance 2.0](https://seedrouter.ai/models/seedance-2-0), [Seedance 2.0 Fast](https://seedrouter.ai/models/seedance-2-0-fast), [Seedance 2.0 Mini](https://seedrouter.ai/models/seedance-2-0-mini) and [Seedance 2.5](https://seedrouter.ai/models/seedance-2-5). All four use the same reference workflow here; their available output settings and prices differ. The controls and model documentation show which settings the selected model supports. \[1]\[2]

For the initial scene, the page prefills **720p, 16:9 and 15 seconds**. These are the Hotel Lobby scene settings, not a claim about every Seedance model’s default request.

| Choice                        | Practical effect                                                                  |
| ----------------------------- | --------------------------------------------------------------------------------- |
| Keep the scene preset         | Use the supplied template length and horizontal framing                           |
| Change the model              | Update the request’s model ID, supported options and live estimate                |
| Change resolution or duration | Change the amount of video being generated and its estimated cost                 |
| Replace the motion reference  | Ask for different movement and change the reference duration used in the estimate |

The estimate includes both output seconds and motion-reference seconds. For this preset, that means 15 seconds of output plus the reference clip’s measured 14.92 seconds. Dollar amounts come from the current model rates, so use the [live pricing section](https://seedrouter.ai/tools/hotel-lobby-ai#pricing) instead of a remembered quote. \[3]

There is no separate Hotel Lobby AI subscription. Generations use your prepaid SeedRouter balance, and another generation is another paid request. Failed requests are not charged.

## 4. Generate, preview and download with sound

Once both performer images are ready, review the estimate and start generation. The result appears in the same Playground. The tool then adds the supplied soundtrack to the silent result without generating another video.

Watch the full clip. Check that the characters stay on their intended sides, that their faces remain recognizable and that hands do not merge around the microphone. Listen to the exported version too: putting the track into the file does not guarantee exact lip sync.

The soundtrack is intended for the roughly 15-second preset. Keep the starting duration to align the template and soundtrack lengths.

If the export fails, use **Retry soundtrack export**. That reuses the completed video rather than starting another paid generation. The original generated video is also available separately.

## What should you change when the result is off?

| What you see                           | What to review before another generation                                                        |
| -------------------------------------- | ----------------------------------------------------------------------------------------------- |
| The performers are reversed            | Check the left and right image slots and any role instructions you edited                       |
| Appearance follows the line-art clip   | Keep the instruction that images provide appearance and the video provides movement             |
| One character is hard to recognize     | Use a clearer source image with a visible face and consistent outfit                            |
| The scene changes midway               | Remove conflicting instructions about new locations, cuts or extra people                       |
| The motion is only approximate         | Preview the reference again and simplify the requested action; exact movement is not guaranteed |
| A reference is rejected                | Check the selected model’s media limits and confirm that the URL is reachable                   |
| The finished download has not appeared | Wait for soundtrack export, or retry export using the existing generated clip                   |

The page checks the supported parameter combinations before submission. Media still needs to satisfy the selected model’s requirements; the [Seedance 2.0 reference](https://seedrouter.ai/docs/seedance-2-0) and [Seedance 2.5 reference](https://seedrouter.ai/docs/seedance-2-5) list those constraints. \[1]\[2]

## Can you automate the same scene?

Yes. The Playground’s JSON tab shows the generation request using the selected model and your reference URLs. Follow the [Seedance API guide](https://seedrouter.ai/blog/seedance-api) to submit a task, save its ID and retrieve the result.

The generated-video API and the soundtrack finishing step are separate. Copying the generation JSON alone does not add the original soundtrack. An application implementing the full workflow needs to add its chosen audio track to the completed video too.

Start with the [Hotel Lobby AI tool](https://seedrouter.ai/tools/hotel-lobby-ai) when you want the images, reference, prompt and soundtrack export in one place.

## Common questions

### Is there a free Hotel Lobby AI generation?

The tool uses your SeedRouter balance and shows the live estimate before you generate. Previewing the template is not a paid generation, but the page does not offer unlimited free rendering.

### Will the soundtrack make the lip sync exact?

No. Export adds the track to the completed video; it does not regenerate the mouth movements. Check the exported clip and decide whether the timing works for your scene.

### Can I use my own motion reference?

Yes. Replace the reference in the Playground and check its media requirements. If you change the movement or length, also review the prompt and whether the supplied soundtrack still fits.

## References

1. [SeedRouter: Seedance 2.0 API reference](https://seedrouter.ai/docs/seedance-2-0) — request fields, reference roles and media limits.
2. [SeedRouter: Seedance 2.5 API reference](https://seedrouter.ai/docs/seedance-2-5) — model-specific options and constraints.
3. [SeedRouter: live Hotel Lobby AI pricing](https://seedrouter.ai/tools/hotel-lobby-ai#pricing) — current model rates and the scene estimate.