Image Generation

Last updated 2026-10-02

How to pick an image model and call GenerateImage with the shared ImageOptions.

Picking an image model

Image and audio models are the agent model, picked exactly like chat models — openai@gpt-image-2, gemini@gemini-3.1-flash-image, grok@grok-imagine-image-2.0, openrouter@google/gemini-2.5-flash-image. The one exception is codex: the ChatGPT backend generates from the chat model itself, so there is no separate image model to pick.

Generating

imageAgent, ok := agent.(llmrouter.ImageAgent)
if !ok {
    return fmt.Errorf("%s cannot generate images", agent.Name())
}
result, err := imageAgent.GenerateImage(ctx, prompt, llmrouter.ImageOptions{AspectRatio: "16:9", Size: "1k"})

Options mapping

ImageOptions is one shared shape each adapter maps onto its own wire parameters; unsupported options are dropped rather than rejected. Size sets the short edge, so 16:9 + 1k becomes 1824x1024.

Related

• Image Types: endpoints and result formats per provider
• Model Filters: listing image models with ImageOnly