Supported client methods
The Nano Banana Quickstart and GPT Image 2 examples provide cURL, Python, and JavaScript requests. A library may be able to send a custom HTTP request, but that does not make its authentication, multipart encoding, path construction, retries, response types, or future versions a supported SDK integration.
Supported REST operations
Every operation uses the same Bearer credential convention. Copy an exact current model ID from the model marketplace; the Nano Banana IDs also appear in the model ID reference.
Similar request shapes do not establish SDK support
Nano Banana resembles Google’s GenerateContent structure, but it does not expose every current or future Gemini field, tool, file workflow, authentication mode, streaming method, or SDK feature. GPT Image uses the familiar synchronous Images Generate and Edit shapes, but its public model IDs differ from OpenAI’sgpt-image-2. Unified task-ID generation and editing use the YingTu-specific /v1/images/tasks JSON contract. Those differences affect model validation, Base64 input, multipart encoding, generated types, retries, and result parsing in an OpenAI SDK.
Asynchronous generation and editing with 1–16 source images use /v1/images/tasks. The task envelope is shared, but Nano Banana and GPT Image keep their native result shapes.
A capability is supported only when it appears in the current OpenAPI and the relevant model Parameters page. A successful REST or Try it request does not certify an SDK version.
If a specific SDK is required for production, send its name, version, language, authentication requirement, and required operations to API support. Until a versioned compatibility matrix is published, use the REST examples as the maintained integration path.
Run the Nano Banana Quickstart
Generate and parse a Gemini-style image response.
Choose a GPT Image route
Select Web or VIP before copying a REST example.