zenstory-ai/video-recap-skills

What it solves

This project provides an automated pipeline to transform long videos into concise, narrated recap videos (summaries) in Chinese. It eliminates the need for manual video editing, scriptwriting, and voiceover production by using AI agents to handle the entire creative process from understanding the content to final assembly.

How it works

The system operates as a set of "skills" that can be integrated into AI agent hosts like Claude Code, Codex CLI, OpenCode, or OpenClaw. The workflow follows these stages:

1. Understanding: Uses ASR (Automatic Speech Recognition) and VLM (Vision-Language Models) via the Xiaomi MiMo API to analyze scenes, extract text, and create a content brief.
2. Planning: An agent acts as a director and editor to create a story plan, visual/audio board, and the final narration script.
3. Cutting: The system can either cut the video first and then write the script based on the edited timeline or write the script first and then cut.
4. Voiceover: Generates high-quality narration using Xiaomi MiMo or Fish Audio TTS.
5. Assembly: Combines the edited video, voiceover, background music, and subtitles using ffmpeg to produce the final MP4 file. It can also export a schema-driven draft for further manual refinement in CapCut (剪映).

Who it’s for

Content creators, social media managers, and users who want to quickly generate video summaries or recaps without needing professional editing software or GPU-heavy local hardware.

Highlights

• Zero Local GPU Requirement: All heavy lifting (ASR, VLM, TTS) is handled via APIs, requiring only Python and ffmpeg locally.
• Agent-Driven Workflow: The AI manages the creative decisions, such as POV and main plotlines, rather than just performing a mechanical summary.
• CapCut Integration: Ability to export editable drafts to CapCut for manual fine-tuning.
• Multi-Video Support: Can synthesize a single story from multiple source videos and reuse analysis results from a local material library to save costs.
• Flexible TTS: Supports multiple providers including Xiaomi MiMo and Fish Audio with customizable voice clones.