AI image negative prompt: what to do when the API has no field

A negative prompt is a second text input that lists what an image should not contain. Sume's image API does not have one: POST /v1/images takes a prompt and no negative_prompt field, so the way to keep something out is to describe the picture you do want, or to edit an image with a mask.

The field list comes from the Image API docs page and the request schema in the Sume API reference, read 2026-09-29.

Which fields can steer an image instead?

The request table lists these controls. None of them is a place to write exclusions, but together they narrow what the model can draw. A model accepts only the parameters its catalog entry lists, and a request that sets an unlisted one is rejected with 400 unsupported_parameter.

How do I keep something out of the picture?

Say what should be there. A prompt that names the thing to avoid gives the model that thing to draw. Rewrite the exclusion as a positive description:

• Instead of "no posters on the wall", write "a plain white wall".
• Instead of "no people", write "an empty street at dawn".
• Instead of "not blurry", ask for the look you want, such as "sharp focus, studio lighting", and raise quality for finals.
• Instead of "no extra text", quote the exact words that should appear and nothing else. AI image with text generator covers that.

How do I remove an object from an existing image?

Send the image as a reference and describe the result without the object. On ChatGPT Image 2.5 you can also send a mask_url to mark the region to change. The docs say both ChatGPT Image 2.5 ids support an optional mask_url and up to 16 image references. GPT Image mask edit API shows the request. On edit calls, prefer aspect_ratio: "auto" to match the reference; omitting the field is not the same as auto.

Can a seed make the unwanted thing go away?

No. seed is in the request schema, but no model advertises it yet, so sending one returns 400 unsupported_parameter. AI image seed explains how to repeat an image with a reference instead. Video has its own answer in AI video negative prompt.

What are the limits?

What the contract does and does not allow:

• The schema closes the request object (additionalProperties: false), so do not invent a field such as negative_prompt.
• Reference and mask URLs must be public HTTPS. Localhost, private-network, and non-HTTPS URLs are rejected before submission.
• Models whose input_references descriptor is {"min": 0, "max": 0} are text-to-image only and reject references.
• Read a model's supported_parameters from GET /v1/images/models before you pin a field.