Vidu Video and Image Generation

Problem It Addresses

Image and video generation workflows often require manual handling of async API tasks, image uploads, reference subjects, and error triage. This skill wraps Vidu's vidu-cli in a CLI-first execution model: commands use flags rather than raw JSON bodies.

How It Works

Core capabilities include:
- Text-to-image / text-to-video: generate media from text prompts
- Image-to-video / start-end frame video: constrain video output with one image or start/end frames
- Reference-based image/video: combine 1–7 images or materials with a required text prompt
- Reference management: create, list, and search elements
- Async task query: submit a task, then poll with task get or task sse

The workflow is: submit, store task_id and trace_id, query until terminal state, then return media_urls or exact error fields. Both success and failure use a stdout JSON contract, which makes scripting straightforward.

Boundaries

It requires vidu-cli >= 0.2.0 and VIDU_TOKEN; it cannot run if the CLI is unavailable. Prompts, images, and task settings are sent to Vidu's API, so privacy and IP requirements must be confirmed. Use the bundled references/parameters.md for task types, model versions, and exact parameter rules.