Vidu Video Generation

Problem

Video-generation scripts often juggle multiple task shapes: text-to-video, a single image as the first frame, multi-subject reference, and first-last-frame transitions, while also managing model duration, resolution, audio-video sync, and async polling. Calling the API directly can lead to wrong model routing, missing image requirements, or ignored download windows.

How It Works

This skill wraps Vidu video generation with a natural-language entry point. After the user provides a prompt or images, it classifies the task and selects an appropriate model. Text-to-video can use models such as viduq3-pro-fast, viduq3-turbo, and viduq2; image-to-video requires one image; reference-to-video supports multiple images; first-last-frame video requires both start and end frames.

Key steps include:
- verifying VIDU_API_KEY
- choosing api.vidu.cn or api.vidu.com based on user language
- submitting the asynchronous task and polling status
- returning Markdown file references and the creations[0].url video link

Boundaries

This is best for quick script validation, sample clips, and draft assets, not for unattended bulk production. Input images should stay under 50MB, generated links expire within 24 hours, and errors such as Invalid API key, Image size exceeds, and Task failed should be resolved by checking the key, asset size, and error details.