Inspiration

Retail teams often need the same product adapted for a marketplace listing, a social post, and a storefront. SceneShift tests whether one clean packshot can cover those placements without another shoot.

What it does

SceneShift verifies the uploaded packshot by SHA-256, then loads an audited batch of Marketplace, Social, and Storefront images. Users can compare the source with each accepted result and download the selected PNG. Each accepted result is a 1024x1024 image.

The batch came from five Perfect Corp AI Photo Background Change API runs. Three passed visual QC, two were rejected, and the accepted set used 20 free promotional units.

How we built it

The demo uses a Python standard-library server and one HTML, CSS, and JavaScript interface. A live adapter calls Perfect Corp's background-change endpoint only when an explicit environment flag and API key are present. This submission uses the sealed batch so judges see the exact inputs and outputs that passed QC without spending more API units.

Challenges we ran into

Two Social attempts changed the bottle body from white to black. We rejected them and tightened the prompt. The final Social image kept the bottle white against a cream background.

Accomplishments that we're proud of

The final batch keeps the white bottle and red band recognizable across all three channels. The sealed upload check also prevents the demo from presenting a different source image as the verified batch.

What we learned

A successful API response is not enough for retail imagery. Product color and silhouette still need review before an image is usable.

What's next for SceneShift

Add channel safe zones and automatic checks for color or silhouette drift before a result reaches review.

Built With

Share this project:

Updates