Hotel Lobby AI videos with two original anime characters

Start with one clear image of each original anime character and assign the left and right parts. Give them a short exchange that you can describe. Keep appearance, action and camera directions in separate parts of the prompt, then watch the whole result against those instructions. The opening frame won't show you how the scene holds together.

Anime is one of Video Hotel Lobby's planned routes, alongside real-person photos whose API integration is in progress. The three existing Seedance entries each have an Anime only label. You can watch our real six-second anime duet and prepare your cast now. Personalized generation, live task quotes and checkout are unavailable.

Choose a scene for the two characters

Choose a moment that shows how the characters relate to each other. One might lead while the other reacts, they might trade replies, or they might finish with a shared gesture. A short exchange gives you a scene to describe and a result to look for. Several unrelated actions are harder to follow.

The difference can come from personality. Put an expressive performer beside a reserved listener, or let a playful partner answer a confident lead. These are ideas for a scene, not tested recipes. Pick an exchange that suits how your characters already behave.

The orange studio and central microphone make the format recognizable, with the two figures standing apart. You can write an original performance for that setting without copying an artist's routine, voice or recording. Use character art and audio you have permission to process and share.

Choose the character references

Choose references with the outfits you want in the video and one intended character in each image. If you need to crop or convert a file, keep the original as well as the prepared copy. Open the file and inspect the detail; large dimensions alone won't tell you whether a face or accessory is clear.

Add a short note to each image. For a fictional pair, that could be "Left: short blue hair, cream jacket, round brooch" and "Right: long dark hair, rust shirt, narrow scarf." Describe what the reference actually shows. Asking for the visible jacket in one sentence and a different jacket in the next leaves the clothing instruction unclear.

Use the image preparation checklist for file formats, visible detail and obstructions. When you look at the two images together, you should be able to tell which character each one is meant to represent.

Assign the left and right parts

The local Video Hotel Lobby preparation screen labels the left reference @Image1 and the right @Image2. Check the previews against those labels and your prompt. They set the intended roles; you still need to watch the output for an unexpected position swap.

Use the same role names throughout the action. If the brief starts with "left character" and "right character," keep those terms instead of switching between names, image numbers and clothing colors. For similar-looking characters, note the visible differences you want to keep and check them during turns or overlaps.

Leave crossing movements out of a first scene unless they are part of the story. Keeping the characters on their assigned sides gives you fewer movements to describe and makes a role swap easier to notice. Treat this as a simpler shot to review, rather than a proven improvement in generation quality.

Write a short exchange

Give the scene a start, a handover and an ending. Choose one main gesture for each character and keep the rest of the movement relaxed. The listener could give a small nod while the speaker gestures, instead of both doing the same dance.

Example scene brief, not a tested prompt: the left character speaks a short line toward the center microphone and makes one open-hand gesture. The right character listens, then answers with a small gesture of their own. Both stay in their assigned positions and finish looking toward the camera.

Adapt the example to the controls your model supports. Written directions won't carry the exact timing of another performance. If you need that timing, check whether the service accepts a motion-reference video you have permission to use and whether the selected mode supports two performers.

The tool selection guide explains the different approaches. Our demo used text to direct an original performance. It had no reference video supplying the original artists' choreography.

Choose the framing

Keep a full-body shot when footwear, posture or body movement matters to the design. For expressions and a spoken exchange, an upper-body shot may suit the scene. Decide which one you want, then watch whether the video keeps that framing all the way through.

Leave space between the figures and think about where the microphone sits. A long ribbon, wide sleeve or hand moving toward the center needs room. Write down how the camera should behave. A tight close-up and two complete figures filling the frame are competing instructions.

Our anime case study asked for both complete figures and allowed a gentle camera push without cropping. The push followed the request, but it eventually cut off the shoes. Simplifying the camera instruction would be a reasonable next test; we haven't tried it. Check the framing even when the characters still look right.

Write the audio plan

Decide whether you want speech, instrumental backing or a scene you'll edit to music afterward. Keep spoken lines short enough for the intended exchange and use original or authorized material. The Hotel Lobby visual setting doesn't include the original song automatically.

Listen for the line you asked for and the voice you intended, then compare the mouth movements with the sound. The presence of an audio track won't answer those questions. Our demo contains audio and played through, but its existing review didn't independently assess the spoken words or precise lip sync.

If you add music later, place it deliberately and watch the exported edit. Shifting the track changes where a beat lands, but it doesn't regenerate the mouths. Keep the original clip so you can compare it with your edit.

Check the model labels and task cost

The three model entries here are Seedance 2.0 (Anime only), Seedance 2.0 Mini (Anime only) and Seedance 2.5 (Anime only). Each Anime only label restricts the intended inputs for that model entry here. The labels don't describe every capability of the suppliers. The separate demo hasn't opened Seedance 2.5 settings or customer generation.

The pricing page lists one-time packs and monthly subscriptions. Purchased credits don't expire. Unused subscription credits expire without rollover, and there are no free personalized generations or signup credits. To price a particular clip, you need the accepted model, settings and full task quote. A pack's price alone won't tell you that cost.

Once generation opens, decide how many attempts you're willing to pay for. Check the quote again when you change settings, and read the failure and refund terms before submitting. A completed clip that misses a creative detail may still be charged; don't budget on an assumed free retry.

Watch the whole performance

Watch the exchange once, then follow each character through the beginning, handover and ending. Compare hair, clothes and accessories as they move. Check body shape, position and framing too. Listen on another pass so a distracting visual detail doesn't make you miss an audio problem.

• Can you identify both characters when they move?
• Does the lead-and-reaction exchange belong to the intended roles?
• Does the chosen crop remain deliberate throughout?
• Is the actual audio suitable for the planned use?
• Does the downloaded file play completely?

Write down what you'd keep and what needs changing. For another attempt, adjust a specific part of the reference or brief and keep the earlier version for comparison. The main Hotel Lobby AI guide covers the full process, and the identity, framing and audio checks help with individual problems. While personalized generation is unavailable, prepare the two references and use the demo to work out your own review checklist.