Generate one image from a prompt. data[0].url is a TextCortex resource URI (textcortex://images/<image_id>), not a directly downloadable HTTP URL. Keep data[0].image_id to reuse the image in another generation. Check GET /capabilities for access. This is a separate image catalog: GET /models lists chat models, not the image models below.

Choose a model

Set model explicitly for predictable selection. Omit it to use gpt-image-2.5 on this deployment.

The table lists presets, not every custom dimension supported by a provider. Use one sizing control at a time: size for pixel dimensions or aspect_ratio for a ratio. Omit both for the model default. For models supporting image_size, combine it with aspect_ratio to select resolution. Unsupported values return 400.

Use a reference image

Add image_urls containing publicly reachable HTTP(S) image URLs, or reuse image_ids returned by a previous generation. File-upload IDs are not image IDs. The model's reference limit applies to both arrays combined. Images must be accessible to the caller; private-network URLs are rejected. Reference URLs must serve JPEG, PNG or WEBP bytes, with a maximum download size of 6 MiB per image. GIF and SVG are not supported.

Each request returns one image: n must be 1 and response_format must be url. Base64 output is not supported. The url response format names the response field; it does not promise a public download URL or a retention period.