One product, three worlds: a match-cut campaign
Design an original product, restage it in two contrasting worlds, then turn the three connected images into a nine-second vertical campaign.
Run it
AI Agent
Terminal
New here? Quick start
brew install pipe2-ai/tap/pipe2 go install github.com/pipe2-ai/pipe2-cli/cmd/pipe2@latest echo "$PAT" | pipe2 auth login --token - Change the result
| Input | Type | default | Description |
|---|---|---|---|
--product | string | defaultone unbranded cobalt-blue sculptural perfume bottle, a squat rounded rectangular opaque glass body with a single large polished silver sphere cap | An original product concept with a simple, distinctive silhouette; avoid labels and tiny markings. |
--scene-one | string | defaulton a pale peach sand dune, surrounded by three enormous curved cobalt-blue petals under a warm apricot sky | The immediately visible opening set. |
--scene-two | string | defaulton a flat mirror-black stone in a lush emerald forest, surrounded by giant wet fern fronds and soft drifting mist | A contrasting environment, preserving the original product and composition. |
--scene-three | string | defaulton a white salt-flat pedestal above a sea of clouds at blue hour, with one huge luminous pale moon behind it | The most expansive final set, held clearly for the payoff. |
-
Designs the product in its opening campaign set, establishing the silhouette, camera and framing.
-
Uses the original image to place the same product in the second set without redesigning it.
-
Returns to the original product reference for the final set, avoiding accumulated changes from successive edits.
-
Animates the three ordered references with hard match cuts, subtle environmental motion and an unhurried final hold. The campaign is intentionally silent.
Why it works and what to change
Why it works
A repeated silhouette gives the eye somewhere to land when the environment changes. The contrast comes from the sets, not from turning your product into something else. Choose three worlds with different colors, textures and scale, then keep the camera and product placement consistent.
GPT Image 2 establishes an original design and its first composition. Nano Banana 2 edits that reference for each new set. Both edits start from the original image: the third scene does not inherit accidental changes from the second. Seedance 2.5 receives all three images as ordered references for a nine-second, 720p vertical campaign. These references guide the shots; they are not exact-frame guarantees.
The default bottle moves from peach dunes to an emerald forest and finally a moonlit cloudscape. This is an imaginary concept campaign, not evidence that a real product was photographed in those places.
The knobs that matter
--product— use a distinctive but simple shape. Fine print, complex closures and transparent internals make consistency harder to judge.--scene-one— make the opening understandable without an introduction. Keep the product visible immediately.--scene-three— give the final set the strongest sense of space or contrast. A final hold should reward watching the sequence.
Use the Run It panel for the complete input list, including the middle set.
Notes
Check the cap, edges and proportions across every cut. Reference-guided generation can still redesign small details, so do not present an approximate concept as an exact commercial packshot. Keep all three scene descriptions compatible with the same product placement.
The result is intentionally silent. Add properly licensed audio when preparing your post, and disclose that the campaign is AI-generated. This recipe does not promise a seamless loop or frame-identical results on reruns.