joeseesun/qiaomu-cut-skill

What it solves

qiaomu-cut transforms a simple text prompt into a complete, reproducible video project. It eliminates the manual effort of searching for stock footage, managing copyrights, configuring ffmpeg, writing subtitles, and adjusting transitions, turning the professional video production process into an engineering workflow that an AI agent can execute and verify.

How it works

The system operates as an "agent-native video director" using a structured approach:

1. Planning: It converts a user request into a QiaoCut IR (Intermediate Representation), which defines the duration, aspect ratio, shot list, and asset strategy.
2. Sourcing: It routes requests to various engines for assets: ClipSeek/Pexels/Pixabay for stock footage, 33tc for movie lines, ListenHub/MarsWave for AI-generated images, video, TTS, and music, and local files.
3. Design: It applies professional cinematic techniques like J/L cuts, push-pull-pan movements, and three-layer bilingual ASS captions.
4. Rendering: It uses ffmpeg-full to synthesize the final product, offering three performance profiles: preview (fast, low res), standard (balanced), and final (high quality, full validation).
5. Verification: It generates render reports and contact sheets to ensure technical quality and license provenance.

Who it’s for

Content creators, educators, and AI agents who need to produce high-quality short-form videos (e.g., English learning videos, product launches, cinematic shorts, or data stories) without manually handling the complex technical pipeline of video editing.

Highlights

• Multi-Engine Routing: Integrates multiple AI and stock asset providers (ListenHub, ClipSeek, 33tc) into a single workflow.
• Professional Typography: Supports three-layer bilingual subtitles with mobile-safe areas and platform-specific font handling.
• Deterministic Rendering: Uses a reproducible IR format to ensure the same project can be rendered consistently across environments.
• Tiered Workflow: Provides a preview $\rightarrow$ standard $>\text{final}$ pipeline to speed up iteration and reduce rendering costs.
• Evidence-Backed Provenance: Automatically generates asset manifests and license reports to track the origin and rights of every clip used.