Pinterest Visual Search Ads: product photo variants from one image

Pinterest Visual Search Ads are ads placed in Pinterest search results and Pin closeups, and visual search starts from a picture, so one product photo is rarely enough creative. With Sume you can send one reference image to POST /v1/images and ask for several angle or context variants, up to 10 images per call with n (per-model ceilings are lower).

Pinterest facts are from its newsroom post, read 2026-10-01. Sume facts are from the Image API docs. This post does not say what Pinterest will accept or rank; check Pinterest's own ad requirements.

What did Pinterest announce?

At Pinterest Presents, Pinterest announced Visual Search Ads. The post says the format places advertisers in Pinterest Search Results and Pin closeups, pairs keyword relevance with a new ad format, and will be available in beta for eligible advertisers in all Pinterest ad markets in the coming weeks. It also says Pinterest users run more than 80 billion searches per month, the vast majority visual.

Why would one product photo need variants?

Visual search matches on what the picture shows, so a single front-on packshot covers a narrow slice of queries. Variants that change the angle, the surface or the room give you more distinct pictures to test. Pinterest's post also mentions self-serve A/B testing in Performance+, which is the place to compare them.

How do I request variants from one reference?

Put the photo in input_references as a public HTTPS image_url, describe the change in prompt, and set n. The docs say reference URLs must be public HTTPS; localhost, private-network and non-HTTPS URLs are rejected before submission.

const res = await fetch("https://api.sume.com/v1/images", {
  method: "POST",
  headers: {
    Authorization: "Bearer " + process.env.SUME_API_KEY,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    model: "openai/gpt-image-2",
    prompt: "same product on a walnut shelf, three-quarter angle, soft window light",
    n: 4,
    input_references: [
      { type: "image_url", image_url: { url: "https://example.com/product.jpg" } },
    ],
  }),
});
console.log(res.status, await res.json());

What limits should I check first?

What should I do next?

List models with GET /v1/images/models, pick one whose input_references range is above zero, and generate a small batch. Keep the product itself consistent by reviewing every variant before it goes into an ad. For a fuller pipeline see AI product photography API.