video-generation-skill

Creates and submits Lingying video generation jobs with duration, reference-image, and provider-state safeguards.

Updated Aug 12, 2026
One-click install
npx skills add https://github.com/VANTSU-1874/woods-video-studio --skill video-generation-skill-vantsu-1874
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: video-generation-skill
Source: https://github.com/VANTSU-1874/woods-video-studio/tree/main/template/.agents/skills/video-generation-skill
Command: npx skills add https://github.com/VANTSU-1874/woods-video-studio --skill video-generation-skill-vantsu-1874

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Producing short-drama episodes requires submitting many per-shot video generation jobs to the Lingying provider, each with strict duration limits, reference images, voice references, and tracked job state. Doing this manually risks oversized segments, missing reference images, lost task IDs, and inconsistent completion tracking. ## Core Features & Use Cases - Job creation and submission: Builds video jobs from episode prompt files (02-视频提示词.md or storyboard 04-故事板视频提示词.md) into the unified jobs system under .claude/production/jobs/, dispatching each job to a media-worker subagent. - Duration and dependency safeguards: Enforces a hard 15-second per-segment limit, requires asset generation and snapshot gates to pass, and blocks storyboard-mode prompts until all storyboard images are completed. - Reference media and voice handling: Builds and validates media arguments so every job carries non-empty reference images, correct model parameters, and per-line dialogue voice references with generate_audio preserved. - Use Case: After finishing the storyboard for episode 1, ask the agent to generate all videos for ep01; it creates one tracked job per shot, submits them through the client tool, and later syncs saved MP4 files with sha256 records into the manifest. ## Quick Start Ask the agent to generate all video segments for episode ep01 from the current episode prompt files and report submission status.

Frequently Asked Questions about video-generation-skill

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I generate videos for an episode from storyboard prompts?

Run the video creation entry point such as `node .claude/tools/ly.js video epXX`, which reads the episode prompt file and creates one job per segment. Storyboard-mode prompts are only used when all non-empty-shot storyboard images for that episode are completed.

What is the maximum video duration per segment?

Each video job must have `durationSeconds` of 15 or less. Longer segments are blocking issues: they must be split or rewritten upstream in the storyboard rather than truncated or submitted as-is.

Can video generation run without reference images?

No. The media arguments must include a non-empty images array, and video generation must not degrade to text-only mode. Reference images come from the episode's locked asset snapshot as immutable copies.

Why did my video job fail with an Invalid URL or ENOENT file error?

This indicates a malformed reference-image parameter, such as a local path converted to a file:// URL. Rebuild the media invocation and retry instead of marking the job failed.

How is video completion confirmed?

A job is completed only when it moves to the completed jobs folder, the local video file exists, output.localPath points to it, a sha256 is recorded, and the manifest counts match the job state.

What happens when regenerating an existing video segment?

The old official video must first be archived into a `历史版本/` folder in the same directory, then the new job is submitted with `save_conflict: overwrite`.