Interface ImageEditCapability<TInput, TOutput>

Image editing capability (non-streaming).

Represents the ability to perform image edits using one or more reference images and an edit prompt.

Image editing is modeled as a first-class capability, distinct from image generation, but shares the same normalization surface (NormalizedImage[]).

Semantics:

• One or more reference images are provided via the request input
• Reference image roles (e.g. "reference", "mask", "style") determine how the provider interprets them
• The provider is responsible for mapping these semantics to its underlying API (e.g. OpenAI Responses API image editing)

This method:

• Executes a single edit operation
• Returns the fully materialized edited image(s)
• Does not stream intermediate results