qwen-video_skill

Overview

This skill provides command-line tooling to generate short videos using Alibaba Cloud DashScope Wan (通义万相) text-to-video APIs. It automates submitting an asynchronous job, polling until completion, and downloading the resulting MP4 to a local path (including Windows Desktop from WSL2). The scripts support model-specific features like multi-shot narration, auto audio, custom audio, silent output, and negative prompts.

How this skill works

You set your DashScope API key as an environment variable and run a submit script to POST a job to the DashScope t2v endpoint. The skill prints the task_id, then a poll script repeatedly GETs the task status until SUCCEEDED or FAILED. When the task succeeds the workflow extracts the video URL and downloads the MP4 to the specified local path, printing MEDIA with the saved file path.

When to use it

• Create a short promotional or social video from a text prompt quickly.
• Produce videos from WSL2 and save output directly to Windows Desktop or other mounted paths.
• Leverage model features like multi-shot storytelling or automatic background audio.
• Run fully automated asynchronous video generation from CI or local scripts.
• Test different models and parameters (size, duration, negative prompts) without building custom clients.

Best practices

• Set DASHSCOPE_API_KEY in your environment before running scripts to avoid auth failures.
• Choose model according to required features: wan2.6 for multi-shot and auto-audio; wan2.2/wanx2.1 for silent output.
• Start with short durations (2–10s) and moderate resolution (832*480 or 1280*720) to reduce cost and turnaround time.
• Use prompt_extend and shot_type: "multi" only with wan2.6 series for multi-shot narratives.
• Provide input.audio_url when you need precise audio control; omit it to let the model auto-generate matching audio.

Example use cases

• Generate a 4s cinematic cyberpunk scene and save MP4 to Windows Desktop from WSL2.
• Batch-generate short ad clips by submitting tasks programmatically and polling for results.
• Create multi-shot teaser sequences using wan2.6 with prompt_extend and shot_type set to multi.
• Produce silent background visuals for installations using wan2.2 or wanx2.1.
• Test negative_prompt to exclude unwanted elements (e.g., remove flowers or logos).