Ideogram 4 describe image to JSON prompt vs Sume reference input

Ideogram's describe endpoint turns an uploaded image into a structured json_prompt that can feed its 4.0 generate endpoints. The Sume docs read for this post list no image-to-structured-prompt route; the Sume way to reuse an image is to pass it as a reference with a prompt.

Ideogram facts are from its describe page, read 2026-10-01; Sume facts are from the image docs.

What does Ideogram's describe endpoint return?

POST /v2/image/describe/ideogram-4 takes a multipart image (max 10MB; JPEG, PNG, and WebP). It returns a description_id, a created time, and a json_prompt of type V4JsonPrompt. include_bbox defaults to true so the prompt keeps the layout; include_style_descriptions and include_tags default to false.

What does Sume offer instead?

Sume's image request takes input_references, described as "Reference images for image-to-image" (image docs). The Image 1.0 page shows the edit shape as prompt plus image_urls with 1 to 10 public HTTPS URLs (Image 1.0). That reuses the picture itself, not a JSON description of it.

What happens if I send an unlisted field?

A request that sets a parameter the selected model does not list is rejected with 400 unsupported_parameter rather than silently dropped. So do not send include_bbox or similar fields to a Sume image route; they are not in the docs read. More on this in GPT Image 2.5 unsupported parameter 400.

What must the reference URL satisfy?

Reference URLs must be public HTTPS. Localhost, private-network, and non-HTTPS URLs are rejected before submission, so there is no multipart upload step on the Sume side; host the file and send its URL.