Water splash, sunscreen, formula texture: heroes from a packshot

Sume's first-party Format catalog includes three beauty hero image Formats that start from an attached packshot and produce a 9:16 still: sume-water-splash-hero, sume-sunscreen-splash and sume-formula-texture-hero. Each is image-only and uses ChatGPT Image 2 per its own instruction. If you sell skincare, they give you three distinct hero looks from a single product photo, to try before you commission a shoot.

What each Format makes

The descriptions below are from the first-party Format definitions in the Sume repository, and the output kind is image for all three. They are examples that give a direction, not guarantees of a specific result.

Calling one

A Format run is POST /v1/formats/sume/{slug}/runs with a key that has the formats:write scope; the input is a free-form object. See the Formats catalog for how to read a Format's io profile before you call it. Bulk runs let you queue up to 100 items, and a run can carry a generation spend cap.

Send a clean packshot against a plain background. The instructions ask for accurate packaging, but the model is redrawing the product, so compare the label text on the result to the real label.

• Pick the Format by the product: water for hydrating, splash for SPF, texture for serums.
• Check the label and cap shape on every output.
• Use the output as a still, or animate it afterward with an image-to-video model.

Limits

Image generation can alter small label text, and a hero that shows the wrong claim is a bigger problem than a flat photo. Marketplace image rules, such as Amazon's requirement for plain-background main images, may not accept these heroes as a main image. Use them for secondary images and ads, and read the rules for each channel.