AI Creative OS
01Make three references, not two02Wire references, skip the start frame03Write the prompt as a shot list04Name the sound, then estimate05QC like an editor, then post the breakdown

The Epic Ad Prompt

How the KAIRO rooftop commercial was made with no actress, no location and no product: three reference images handed straight to Seedance 2.5, one timed prompt, all through one connector. The exact prompt is below, ready to paste.
What you'll make
Build it step by stepInteractive · ~4 min · with Z’s build guideStart →
01Step 1
Make three references, not two
Open Nano Banana Pro (gemini-3-pro-image) and make three stills. One: the product on seamless white with the wordmark in quotes in your prompt so it renders exactly (for a client, use their real packshot). Two: a portrait of the lead, written like a casting note: hair, features, jacket, the light on her face. Three: the location and the look, a single still of the world the spot lives in (here: a rain-soaked rooftop, magenta and teal neon, wet gravel, mist). Add the realism layer every time: visible pores, flyaway hairs, no smoothing, film grain, muted grade.
02Step 2
Wire references, skip the start frame
In Seedance 2.5, connect all three stills to reference_images. The model refuses references together with a start frame, so do not add one. That is the difference from the Kling runs: Kling starts from one locked frame, Seedance builds every shot from the references, which is what lets it change angle, scale and lighting inside one clip. Open the prompt by saying what each reference is for: the lead, the product, the location and look.
03Step 3
Write the prompt as a shot list
Copy the prompt below and change the product, the lead, the world and the four beats. Keep the shape: an identity and product lock at the top, four timed shots (arrival, the product, the use, the hero wide), each with one camera move and one action, then a style line and a negative line. Stay under 300 words. One slow-motion moment is enough; put it on the product.
04Step 4
Name the sound, then estimate
Turn audio on and name every sound you want: rain on leather, the tab hiss, distant thunder, a rising swell, no dialogue. It comes out of the model, so there is no sound library and no music license. In Claude Code, the ElevenLabs connector runs the image model and the video model behind one login (add it with claude mcp add --transport http elevenlabs https://api.elevenlabs.io/v1/mcp). Ask for the estimate first: Seedance 2.5 at 1080p is roughly $0.80 a second, 720p about half.
05Step 5
QC like an editor, then post the breakdown
Watch the clip three times: once for her face across the cuts, once for the can (same wordmark, same ring, same lid), once for the hands. If the label drifts, fix the packshot and run again, never re-roll blind. Then post the breakdown, not just the spot: the clip on top, the references under it, the prompt scrolling while it plays. For a client, the three references are the brief: change them and you change the brand, the lead and the world.
The exact Seedance 2.5 prompt (copy it, change the product, the lead, the world and the four beats)
Create a cinematic 15-second energy drink commercial using the provided woman as the
consistent lead and the provided KAIRO can as the exact product reference. Keep her
face, sleek black high ponytail, silver ear ring, black leather motorcycle jacket and
grey hoodie identical in every shot. The can keeps its exact shape, matte black body,
vertical copper KAIRO wordmark, thin copper ring and silver lid. Do not redesign or
relabel it. The third reference is the location and the look: a rain-soaked city
rooftop at night, magenta and teal neon, wet gravel, mist.

0:00-0:04 ARRIVAL. Low wide shot on the wet rooftop. She walks toward the lens out of
the mist, helmet under one arm, neon reflecting in the puddles, slow push-in. Thunder
far away, rain ticking on leather.

0:04-0:08 THE CAN. Macro close-up, shallow depth of field. Her gloved hand lifts the
KAIRO can into frame, condensation running down the matte black, copper wordmark
catching the teal light. Her thumb cracks the tab in slow motion: a sharp hiss, a
fine burst of cold mist.

0:08-0:12 THE SIP. Medium close-up, camera orbits a quarter turn around her. She
drinks, lowers the can, exhales a breath of vapor, eyes opening toward the skyline as
every neon sign behind her flares brighter.

0:12-0:15 HERO. Fast pull-back to a wide skyline shot: she stands at the roof edge,
can in hand at her side, city glowing below, lightning sheet in the clouds. Hold.

Style: anamorphic lens, shallow depth of field, real rain, wet reflections, film
grain, muted cinematic grade with teal and magenta practicals, realistic skin.
Audio: rain, distant thunder, the can hiss, a low rising synth swell, no dialogue.
No morphing hands, no label changes, no extra people, no on-screen text.
Want this running for you?
Build it step by step inside AI Creative OS, with me in the thread.
Implement this today →
Zdeno Mucina
Not ready for that?
Drop your email and 3 more guides open in the Library, free.