Baiyin Video Generation

Problem

When calling the Baiyin video generation API, engineers often juggle models, generation types, duration, resolution, aspect ratio, media URLs, and prompts. Missing parameters, submitting local paths, or consuming credits without a final confirmation are common failure modes. This skill turns that scattered exchange into a fixed workflow: inspect models and parameters, collect media, confirm configuration, then submit and poll.

How It Works

• Models and parameters: silently reads BAIYIN_API_KEY and fetches available models; asks the user when multiple models or types apply, and silently applies singleton values.
• Media collection: collects URLs for text-to-video, image-to-video, first/last frame, multi-image reference, audio-driven, and video-reference modes; local paths trigger upload and replacement with public URLs, with a maximum of 6 images for multi-image reference.
• Prompt handling: treats prompts as required, optional, or skipped based on the selected scheme; offers a polished version when the prompt is thin, while keeping prompt selection separate from parameters.
• Submission and polling: shows a configuration summary and requests confirmation; after submission, displays taskId and requestId, polls every 30 seconds until succeeded, failed, or a 10-minute timeout.

Boundaries

This skill fits interactive use of existing Baiyin video generation capabilities. It does not replace model inference or guarantee task success; if required fields cannot be supplied, choose another scheme. Confirmation is mandatory before submission, and timeout requires manual lookup using taskId.