Make Your Own Alors On Danse AI Duo Clip in 10 Minutes (Seedance 2.5)

Seen those clean white-studio duo skits? One person calmly taps a mini keyboard while the other hypes them up—then the partner leans right into the camera. You can make one with a friend (or two pets) that actually looks like you.

This guide is for first-timers. Follow the steps below in the APIPod Seedance 2.5 web Playground—the page where you upload images, fill a few fields, and get a clip—and you can have a vertical short in about 10 minutes. No code required. If you later want a small tool or batch runs, there’s an optional API note at the end.

30-second brief: what this is—and what to watch for#

The look: a clean white studio, one table, a mini keyboard; a steadier “keyboard player” on the left and a louder “hype partner” on the right; clear beat-driven beats; often a lean-in at the end. Vertical 9:16 is the sweet spot for TikTok / Douyin.

A few friendly boundaries before you start:

• Not an official one-tap filter. This isn’t a CapCut official template, and it isn’t the early-2020s “everyone dance together” group challenge.
• Where the visual meme comes from. These white-studio duos sit closer to a jokey “coach directing the keyboard player” bit from Made in Jamel (2010) than to a real studio session behind some famous track. The AI trend just swaps that calm-vs-hyped pair for you and a friend—or two pets.
• Get consent. For real people on camera, both sides should clearly agree. This post only covers “you and a friend / two pets”—don’t try it with unauthorized celebrity photos.
• Music rights are separate. The model gives you picture (and optionally a bit of ambient audio). That is not a license to slap a hit song on the cut. For the final score, use a platform music library or audio you’re already cleared to use.

Clear on that? Then you’re free to play.

The frame above locks the casting in one glance: left steady, right loud; keys first, lean-in last. When you upload photos and write the prompt, don’t swap left and right by accident.

Before you click: checklist + photo tips#

You only need a short kit:

When you shoot, match the left/right of the tip sheet below: left = do this; right = try not to.

Concrete rules:

• Do: front or slight three-quarter, face sharp, even light, half-body to the waist line if you can, clean background; clothes that contrast (one calm, one lively) so the model can tell the roles apart.
• Don’t: hands over the face, hard backlight that melts the face, tiny distant full-body shots, heavy beauty filters that warp features, or a collage of many heads in one frame.
• Practical tip: shoot three candidates each and pick the one with sharp eyes, a natural expression, and no motion blur. On vertical, step back half a pace so shoulders and table props have room—this skit lives on “half-body + hands visible.”
• Pets: prefer stills looking up at the camera with clear fur edges. Two pets can play “steady” and “hype”; just say who is left and who is right in the prompt.

Once the kit is ready, skim the end-to-end flow so the buttons feel familiar:

For most people, the first pass is enough: two photos → web clip → edit with licensed music. Only if you want multiple reference images—or a motion reference video—do you switch model routes (plain-language comparison at the end).

Step by step on the web: open Playground and click through#

Open the image-to-video page (photos → video):

Seedance 2.5 Image to Video

On that page, hit Try in Playground. The steps below use the real form field names—match them on screen.

1. Paste the Prompt (required: text instructions for the model)#

Don’t stop at “two people dancing”—that basically asks the model to invent the whole scene. The “How to write the prompt” section below has a layered checklist, Chinese and English starters, and a mistakes table. Paste a starter, change the names, skim the checklist, then hit Run.

2. Upload Start Frame / End Frame#

• Formats: JPG / PNG, roughly up to 10 MB each.
• Start Frame: prefer a duo medium shot, or the clearer keyboard-player three-quarter; left/right must match the prompt.
• End Frame (optional): for a more stable ending lean-in close-up, upload a hype-partner close shot.
• If left and right keep flipping, swap the two images and re-run—often faster than rewriting the whole prompt.

3. Duration: pick 4–30 seconds#

To check faces and roles first, 4–6 seconds is enough. Once that looks right, try 8–12 seconds for a short you can publish. Don’t max out at 20+ seconds on attempt one—longer trials cost more and teach you slower.

4. Resolution: 480p / 720p / 1080p#

Debug at 480p or 720p; only go to 1080p when framing and casting already feel right.

Rough cost (soft guide—live page pricing wins):

Example: 720p × 10 seconds sits around a couple of dollars—still trust the price and billing shown when you submit.

5. Aspect Ratio#

Default is adaptive. For TikTok / Douyin vertical, prefer 9:16 when it’s in the dropdown. The lean-in hit usually lands harder in vertical than in landscape.

6. Generate Audio#

Turn it on if you want a bit of ambient or model-synced sound. Remember: whether you can post that famous song is still on you and your licensed source—don’t confuse “model made sound” with “legal to publish the hit track.” For tight drum hits, text can only bias “follow the beat”; don’t expect the Prompt alone to nail every downbeat.

7. Hit Run#

Submit, wait a bit, then download from the output area. A good take should show a clean white set, matching left/right roles, rhythmic interplay, and ideally one clear lean-in. If v1 is off, the next section shows what to tweak; for prompt-level fixes, jump to “How to write the prompt.”

How to write the prompt: lock relationship, action, and camera in layers#

A weak prompt like “two people dancing” hands casting, camera, white set, and rhythm to pure guesswork. A strong prompt doesn’t need every layer in essay form—but it should at least lock relationship + action + camera. Think in these ten layers (none has to be long):

1. Cast: who appears (you and a friend / two pets); only enough looks to stay recognizable
2. Roles: left = calm keyboard player; right = hyped partner
3. Action: body facing, arms, energy, how the two interact, how it escalates over time
4. Attitude: playful, exaggerated, a touch of comedy—a few sharp words beat a pile of empty adjectives
5. Camera: start medium-wide, then a restrained slow push-in; don’t demand “locked-off + wild follow + constant zoom” at once
6. Framing: both people stay in shot; left keys, right hype—don’t fight for center
7. Light: bright clean studio, soft shadows, consistent start to finish
8. Rhythm: action accents follow the beat; exact lip-sync / drum-grid sync usually needs licensed audio in edit—don’t expect pure text to nail millisecond timing
9. Background: nail “minimal white studio,” or clearly swap the set (living room, stage)—leave it blank and the scene tends to drift
10. Overall style: clean, playful, lightly comic, fluid coherent bodies

Pre-Run layered checklist#

Tick these, then submit—outputs stay steadier:

• Who are the two roles? What does left vs. right play?
• What is each doing? How do they interact, and how does energy change?
• Attitude: two or three concrete words (not ten empty ones)?
• Where does the camera start and push to? Any fighting camera instructions?
• Are both people in frame? Left keys / right hype written in stone?
• Light and background (white studio or the set you want) spelled out?
• Rhythm = “accents on the beat,” not a fantasy that text hits every bar of a specific hit song?

Common mistakes (skip these first)#

中文起步稿（完整版 · 约 8–10 秒）#

纯白极简摄影棚，干净无缝背景，桌前有小键盘。你和朋友（或两只宠物）演双人戏：左侧冷静键盘手，右侧夸张打气搭档；两人始终都在画面里。
0–3 秒：中全景起幅。左侧身体微侧向键盘，手指稳定有节奏敲击；右侧站在旁侧，轻点头观察，肩臂小幅跟拍。
3–7 秒：右侧开始鼓劲——指向键盘、打拍子、上半身夸张但友好的律动手势；左侧仍专注演奏，偶有点头合拍。表演态度：俏皮、夸张、略喜剧，别堆空形容词。
7–10 秒：右侧探身 lean-in 靠近镜头，表情兴奋；镜头从中全景克制缓推到近景，勿乱晃、勿突然变焦打架。灯光明亮均匀、柔阴影全程一致；背景保持白棚。整体肢体自然连贯，节拍重音清楚（精确对某首歌请后期配授权音乐）。竖屏 9:16。

中文起步稿（短版）#

白棚双人：左冷静敲小键盘，右夸张打气。两人始终入画，左键右闹。中全景起，克制缓推；右侧打拍、指键、最后探镜 lean-in。明亮柔光、干净白背景。俏皮略喜剧，动作跟节拍重音走。竖屏 9:16。

English starter (fuller · ~8–10s)#

Minimal clean white photo studio, seamless backdrop, a table with a mini keyboard. Subjects: you and a consenting friend (or two pets). Left = focused keyboard player; right = playful hype partner. Keep both clearly visible.
0–3s: Medium-wide start. Left angled to the keys, steady rhythmic tapping; right watches beside them with light nods and small shoulder accents.
3–7s: Right hypes—points at keys, claps the beat, exaggerated but friendly upper-body dance gestures; left stays focused, occasional head-bob on the beat. Attitude: playful, slightly comedic—few sharp words, not adjective spam.
7–10s: Right leans into the camera; controlled slow push-in from medium-wide to closer framing—no conflicting handheld chaos or constant zoom wars. Bright even studio light, soft shadows, consistent; keep the white set. Fluid coherent motion with beat-synced accents (exact track timing = add licensed audio in edit). Vertical 9:16.

English starter (shorter)#

White-studio duo: left calm on mini keyboard, right energetic hype partner. Both stay in frame; left keys / right hype. Medium-wide start, controlled push-in; right claps, points, then lean-in. Bright soft light, clean white backdrop. Playful, slightly comedic; beat-synced accents. Vertical 9:16.

Swapping cast: change the people line, keep the skeleton#

You can reuse the scene skeleton (white studio, left keys / right hype, medium-wide → push-in, lean-in) and only rewrite the “who is who” lines:

• Friends: spell appearance contrast and left/right roles; keep action and camera as-is.
• Couple: add a line about natural chemistry and exaggerated groove, but keep left steady / right loud—don’t turn it into a pure staring romance film.
• Pets: say “human on keys / pet as hype partner” (or two pets as steady vs. hype); write interaction as “looks up, paws forward, bobbing with the beat”; leave camera and white set alone.

The weirder the pairing, the more you must nail left vs. right in writing.

If you go r2v (reference video)#

With a reference clip, don’t invent a brand-new choreography in the Prompt. Write what to keep: main action and interplay, performance timing, original camera and framing, rhythm and groove, background structure. Text stops the shot from drifting; the reference clip owns “how it moves.”

Suggested iteration order: lock “who is left / who is right” → tune “when the lean-in hits and how hard the push-in is” → only then add wardrobe and mood words. With the same photos, change only the beat marks across two or three runs—that finds the issue faster than rewriting images and text at once.

First take not there yet? Fix by symptom#

A low-stress loop: keep the same photos, only change beat marks in the prompt (e.g. move lean-in from second 7 to second 5), run two or three versions, and keep the one that feels closest.

Finishing the cut: CapCut (or any editor) + licensed music#

The model hands you picture (and optional generated audio). The last mile is yours:

1. Trim: cut dead air at the head or shake at the tail so action lands on the drums.
2. Score: add licensed audio in your editor (platform library, catalog you have rights to, etc.). Hit-song copyright is a separate question—don’t assume “AI made the picture, so I can use any track.”
3. Captions (optional): a short line or role nicknames often helps completion.
4. One more check: on-camera consent, platform rules, and no accidental use of someone else’s likeness.

Even if you’re only DMing a friend for laughs, prefer tracks the platform already clears for posting—that habit keeps public posts safer later.

(Optional) Building a small tool / batch runs: short API note#

Most creators can stop at the web path above. Only if you want a script or many clips at once do you need the API (programmatic access to the same capability).

Remember one sentence: video is an async job—you submit, get a task id (task_id), poll until status is complete, then download. It is not a “click once, get the whole video back immediately” sync API, and it is not OpenAI-video-API compatible.

• Submit: POST https://api.apipod.ai/v1/videos/generations
• Poll: GET https://api.apipod.ai/v1/videos/status/{task_id}
• Auth header: Authorization: Bearer <your API Key>
• Common model names: seedance-2.5-i2v (image-to-video), seedance-2.5-r2v (multi-reference), seedance-2.5-draft (cheap draft, then a sharper pass)

Which model? Plain-language map:

Minimal submit example (swap in publicly reachable image URLs):

curl -sS https://api.apipod.ai/v1/videos/generations \
  -H "Authorization: Bearer $APIPOD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seedance-2.5-i2v",
    "prompt": "Clean white photo studio. Two people at a table: left calmly taps a mini keyboard, right hypes and points, then leans into the camera. Clear rhythm, natural body motion, vertical framing.",
    "image_urls": [
      "https://cdn.example.com/friend-a-start.jpg",
      "https://cdn.example.com/friend-b-end-closeup.jpg"
    ],
    "duration": 10,
    "resolution": "720p",
    "generate_audio": true
  }'

Poll with the task_id from the response:

curl -sS "https://api.apipod.ai/v1/videos/status/vid_task_01JEXAMPLE" \
  -H "Authorization: Bearer $APIPOD_API_KEY"

Common statuses: pending → processing → completed / failed. Only pull the video URL after completed. Deeper params: official docs for Image-to-Video · Reference-to-Video.

FAQ (plain English)#

Do I need two people?
The joke is “calm keyboard player vs. hyped partner.” Two photos are the stable path; one photo usually underperforms—don’t treat that as the main route.

Is web upload the same as image URLs in code?
On the web you can upload local JPG/PNG. Via API you generally need publicly reachable image URLs; private intranet addresses usually fail.

Can generated audio stand in for a hit song when I publish?
Don’t assume that. Score the final cut with music you’ve confirmed you’re cleared to use.

Is this an official CapCut template?
No. This post uses Seedance 2.5 on APIPod to recreate a similar visual trend. The prompts here are starters—you iterate.

Pets? Celebrities?
Pets are fine as a light example. Celebrities or other unauthorized likenesses: don’t use them as your subjects.

Go make your white-studio duo#

Grab two photos you’ve consent for, open the Playground, and click through as above:

• Seedance 2.5 Image-to-Video Playground
• Seedance 2.5 model overview
• When you need programmatic access, get an API Key

Run an 8–10 second vertical pass of “you and a friend” or “two pets” first; only then lengthen and upscale. Have fun—and stay just as serious about consent and music rights. Clearance doesn’t kill creativity; it lets you ship the fun cut without second thoughts.