linzzzzzz/openclip

What it solves

OpenClip is a lightweight automated video processing pipeline designed to identify and extract the most engaging highlights from long videos, such as livestreams or talking-head videos. It eliminates the manual effort of scrubbing through hours of footage to find viral-worthy clips, automatically handling everything from downloading and transcription to AI-driven analysis and final clip generation.

How it works

The system follows a multi-stage pipeline:

1. Ingestion: Downloads videos from Bilibili, YouTube, or local files.
2. Transcription: Uses platform subtitles or local ASR (Whisper for English, Paraformer for Chinese).
3. Analysis: If the video is over 20 minutes, it is split into segments. An LLM (via Qwen, OpenRouter, GLM, MiniMax, or custom OpenAI-compatible APIs) analyzes the content based on engagement and entertainment value.
4. Optimization: Users can enable a deep-optimize mode that adds a secondary AI review and boundary correction phase to ensure clips are independent and natural.
5. Production: Generates independent video clips with accompanying subtitles, summaries, and cover images. It can also burn subtitles into the video and add stylized artistic banners.

Who it’s for

• Content Creators: Those looking to quickly turn long-form livestreams or podcasts into short-form clips for social media.
• AI Agents: Users of Claude Code, TRAE, or Cursor can use it as a skill to process videos via natural language.
• Developers: People wanting a lightweight, customizable alternative to heavy Docker-based video processing tools.

Highlights

• Multi-Interface Support: Available via a Streamlit web UI, CLI, and as an Agent Skill.
• Intelligent Analysis: Supports --user-intent to guide the AI toward specific topics (e.g., "views on AI risk").
• Speaker Identification: A preview feature that labels specific speakers in subtitles using reference audio clips.
• Flexible Post-Processing: Includes subtitle burning with optional LLM-based translation and various artistic banner styles (e.g., neon, metallic, crystal).
• Lightweight Design: Built with Python and FFmpeg, avoiding heavy infrastructure like Redis or PostgreSQL.