Video Generation

Use the native video_generate tool when the user asks you to create, generate, render, or make a deliverable video.

Do not use video_generate for image generation or video understanding. Use image_generate for still images and the appropriate vision/video analysis tooling for inspection.

Workflow

1. Call video_generate with action: "list" if you need to check whether a video provider is configured or explain missing auth/model setup.
2. For generation, pass a clear prompt and optional aspectRatio, resolution, and durationSeconds.
3. The native provider layer supports OpenAI Sora 2 Pro / Sora 2 and Google Veo 3.1 Fast / Veo 3 when those providers are configured.
4. Treat provider warnings as user-relevant when they change requested options such as aspect ratio, resolution, or duration.
5. Return the generated artifact directly in the final response. The tool owns provider auth, provider quirks, file persistence, and media delivery paths.

Keep prompts concrete and cinematic. Include subject, movement, camera angle, duration, aspect ratio, lighting, style, and any required audio direction when the selected provider supports audio.