GPT Image 2.5 multi-turn edits: previous_response_id vs Sume

OpenAI's Responses API carries image edit history for you through previous_response_id. Sume's POST /v1/images has no such field in the docs I read: for the next turn you send the previous result's URL back in input_references with a new prompt.

OpenAI facts are from its image generation guide; Sume facts from the Image API docs, read 2026-10-01.

How does OpenAI do multi-turn image generation?

The guide lists multi-turn editing as a capability and says that, with the Responses API, you build multi-turn image conversations either by providing image generation call outputs in context (you can also just use the image ID) or by using the previous_response_id parameter.

What does one turn look like on Sume?

Each call is its own request. The result is data[].url, a Sume-hosted signed URL. To refine it, pass that URL as an image_url item in input_references. Reference URLs must be public HTTPS, and for edit calls the docs recommend aspect_ratio: "auto" to match the reference. GPT Image 2.5 takes up to 16 references.

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-2.5",
    prompt: "make the background a night sky, keep the subject",
    aspect_ratio: "auto",
    input_references: [
      { type: "image_url", image_url: { url: "https://example.com/turn-1.png" } },
    ],
  }),
});
console.log(res.status);

How do the two approaches compare on state?

What should I watch for?

If a turn exceeds 30 seconds you get 202 and a job envelope, so check the status code and read the result from the job endpoints. Because turns are separate, the signed URL you reuse must still be reachable over public HTTPS. More on chaining edits in multi-round image edits.