alicloud-ai-video-wan-video_skill

Overview

This skill standardizes video generation for the video-agent pipeline using the DashScope Python SDK with the exact model name wan2.6-i2v-flash. It provides a normalized video.generate interface, clear input/output mappings, and practical guidance for async handling, caching, and operational concerns. Use it to implement reliable, reproducible image-to-video synthesis in production or documentation.

How this skill works

The skill maps a normalized video.generate request to DashScope's VideoSynthesis calls, enforcing the required model string wan2.6-i2v-flash and translating reference_image into img_url (the SDK auto-uploads local files). It supports synchronous or asynchronous flows: call, check for task IDs, poll until completion, and return a standardized response with video_url, duration, fps, and seed. Inputs like prompt, duration, fps, size, and optional seed/negative_prompt/motion_strength are validated and passed through.

When to use it

• Integrating image-to-video generation into a video-agent pipeline that needs stable, reproducible outputs
• Documenting or implementing video.generate requests/responses for teams using DashScope
• Needing a canonical mapping from your API fields to DashScope (reference_image → img_url)
• Building asynchronous workflows with task polling, progress reporting, and cancel/retry support
• Caching generated assets by request fingerprint to avoid duplicate renders

Best practices

• Always use the exact model string wan2.6-i2v-flash; do not use aliases or date suffixes
• Provide reference_image (URL or local path) — the model requires an input image
• Implement polling and surface progress; video generation can take minutes
• Cache by full request fingerprint including reference_image hash to prevent duplicate work
• Store final videos in object storage and persist only URLs in metadata

Example use cases

• Backend service converting user-supplied concept art into short promotional videos
• Automated content pipeline that generates thumbnails and preview clips from a reference image
• A documentation example showing exact parameter mapping for prompt, fps, duration, and size
• A video-agent that accepts user prompts and reference images, queues jobs, and streams progress
• Batch rendering where identical requests are deduplicated via cache keys