GPT Image 2 input_fidelity: omit it; Sume returns 400

For gpt-image-2, OpenAI says to omit input_fidelity: the API does not allow changing it, because the model processes every image input at high fidelity automatically. Sume's image request table has no input_fidelity field, and a request that sets a parameter the selected model does not list is rejected with 400 unsupported_parameter.

OpenAI's statement is from its image generation guide; Sume's from Image generation, read 2026-10-01.

What does input_fidelity do on OpenAI?

The guide says the parameter controls how strongly a model preserves details from input images during edits and reference-image workflows. For gpt-image-2 you leave it out, since high fidelity is automatic.

What does Sume do if I send it anyway?

Sume does not silently drop unknown parameters. The docs say a parameter the selected model does not list is rejected with 400 unsupported_parameter rather than silently dropped. If you port OpenAI code, delete input_fidelity before the request leaves your app; do not expect it to be ignored.

What do I send for an edit?

Put reference images in input_references, described in the docs as reference images for image-to-image. The documented request fields include model, prompt, n, resolution, aspect_ratio, quality, output_format, mask_url, background and input_references. The quality field accepts auto, low, medium, high, xhigh or max where the catalog allows it, so quality is the lever Sume lists. Details for editing are in GPT Image 2.5 image editing.

curl -X POST https://api.sume.com/v1/images \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-image-2.5",
    "prompt": "Keep the product, change the background to a studio sweep.",
    "input_references": [
      { "type": "image_url", "image_url": { "url": "https://example.com/product.png" } }
    ],
    "quality": "high"
  }'

How do I port quickly?