SuperAPI AI Directory
Partner with us中文AI newsFavorites
DirectoryNews and tutorials

CutClaw - An AI Video Editing Tool Jointly Open-Sourced by Great Bay University and Beijing Jiaotong University

Author: AI小集 · Original source: AI工具集

What is CutClaw

CutClaw is open-sourced by the GVC Lab of Great Bay University and the Beijing Jiaotong University team.AI video editing tool. The tool adopts a multi-agent architecture and, through a "music-driven" approach, automatically edits hours-long videos into short films with precise rhythm. The system first parses the music's beats and structure, then combines them with the user's text instructions; an AI screenwriter plans the shots, an editor selects the clips, and a reviewer performs quality checks, ultimately rendering a cinematic video suitable for multiple platforms. CutClaw supports one-click material decomposition and cache reuse, making it suitable for scenarios such as travel videography and marketing.

CutClaw

CutClaw's Main Features

  • Music-driven editing: Analyzes music beats, downbeats, and energy curves, strictly aligning visual narration with the music structure to achieve true audio-visual synchronization.
  • Multi-agent collaboration: Simulates a professional post-production workflow: AI screenwriter (plans story rhythm and shots), AI editor (selects clip timings), AI reviewer (quality-checks shot length and aesthetics), forming a closed-loop optimization.
  • Instruction-based control: Only a single text description is needed (e.g., "show the protagonist's madness"), and the system automatically understands the style and executes it, with no need to manually drag the timeline.
  • Intelligent material decomposition: One-click decomposition of hours-long videos into a structured shot library, annotating cinematography techniques, character emotions, and narrative nodes; audio extracts beat and energy features and converts them into searchable assets.
  • Content-aware cropping: Automatically identifies the core subject in the frame and intelligently adjusts the aspect ratio (9:16, 16:9, etc.) to suit multi-platform publishing needs such as Douyin and Xiaohongshu.
  • Cache acceleration: Caches the decomposition results after the first processing, and directly reuses them for subsequent editing of the same material, greatly improving efficiency.

How to Use CutClaw

  • Set up the environment: After cloning the code repository from GitHub, create a Python 3.12 virtual environment and install the project dependencies.
  • Prepare materials: In the resource/ directory, place the video and audio files respectively; optionally place a subtitle file to skip speech recognition.
  • Start running: Execute streamlit run app.py to launch the visual interface, or run directly from the command line by passing file paths and instruction parameters.
  • Configure models: In the configuration file, set the API keys supported by LiteLLM, and separately specify the large models used for video understanding, audio parsing, and agent reasoning.
  • Get the final video: Wait for the system to automatically complete material decomposition, shot planning, and editing rendering, then download video files in multiple aspect ratios suitable for various platforms.

CutClaw's Key Information and Usage Requirements

  • Project background: An AI video editing system jointly open-sourced by the GVC Lab of Great Bay University and Beijing Jiaotong University, implementing music-driven automatic editing of long videos based on a multi-agent architecture.
  • Core mechanism: Adopts a "screenwriter-editor-reviewer" multi-agent pipeline, decomposes materials to generate structured subtitles, plans shots based on music beats (downbeats/energy/pitch), and finally renders a cinematic short film with precise rhythm.
  • Technical dependencies: Calls large model APIs through the LiteLLM gateway; for video understanding, Gemini-3/Qwen3.5 is recommended; for audio parsing, Gemini-3 is recommended; for agent reasoning, MiniMax-2.7/Kimi-2.5 is recommended.
  • Environment configuration: Python 3.12, Conda environment, GPU (CUDA) is strongly recommended for video codec acceleration.
  • File preparation: You need to place videos (.mp4/.mkv) and audio (.mp3/.wav) into the resource/ directory. Optional .srt subtitles can skip ASR to save time and API costs.
  • API configuration: API keys for each model provider (OpenAI, Google, Moonshot, etc.) must be configured, set via environment variables or a configuration file.
  • How to run: Supports the Streamlit visual interface (streamlit run app.py, access localhost:8501) or the CLI command line (python local_run.py pass in path and instruction parameters).

CutClaw's core advantages

  • True music-driven editing Unlike traditional tools that "edit the video first, then add BGM," CutClaw first deeply analyzes the music's beats, downbeats, and energy curve, letting editing decisions be fully driven by the music structure to achieve true audio-visual unity.
  • Professional-grade multi-agent collaboration Simulates the entire film post-production workflow: the AI screenwriter plans the narrative rhythm, the AI editor selects precise clip timestamps, and the AI reviewer performs quality checks (shot length, protagonist proportion, aesthetic score), forming a self-correcting closed loop rather than a one-shot generation.
  • End-to-end long video processing Optimized for the scenario of "cutting hours of footage into a few minutes of short video." One click deconstructs massive footage into structured, searchable assets. Combined with a caching mechanism, it enables an efficient workflow of "slow on the first cut, fast on re-cuts."
  • Zero-threshold instruction control No professional knowledge required. A single natural-language description (such as "show the Joker's madness and elegance") can drive stylized editing, automatically understanding emotion, rhythm, and visual preferences.
  • Native platform adaptation Content-aware smart cropping automatically identifies the main subject in the frame, generating multiple aspect ratio versions in one click, including 9:16 (Douyin), 16:9 (Bilibili), and 1:1 (Xiaohongshu), saying goodbye to black bars and cropping mistakes.

CutClaw's project links

  • GitHub repository：https://github.com/GVCLab/CutClaw
  • arXiv technical paper：https://arxiv.org/pdf/2603.29664

Comparison of CutClaw with similar competitors

Comparison dimension CutClaw OpusClip Mora
Core positioning Long-video cinematic editing, music-driven narrative Long video to short video, viral clip extraction Video generation, multi-agent scene coordination
Music synchronization method First parse the music structure(beats/energy/verses and choruses), then drive visual editing decisions Supports music beat alignment, focusing on highlight extraction followed by adding music Focuses on visual consistency; music synchronization is not a core feature
Long-video support Hours-long(Hours-long) end-to-end processing Supported (podcast/livestream replay to short video) Supports long-sequence generation
Architecture characteristics Multi-agent closed loop(screenwriter + editor + reviewer collaboration) Single-model algorithm recommendation Multi-agent(similar to CutClaw's architecture)
Open source Yes No Yes
Control method Natural language instructions to control style Automatic extraction + manual adjustment of clips Text prompt-controlled generation
Use cases Travel photography/Vlog cinematic production, film and TV fan edits Social media marketing, livestream clip cutting Creative video generation, virtual scene construction

CutClaw use cases

  • Travel photography and Vlog production: Pair hours of travel footage with background music to quickly generate cinematic short videos with precise rhythm and natural beat-syncing, greatly saving post-production time.
  • Film and TV fan edits and mashups: Re-edit movie or series clips based on a specific musical rhythm, automatically generating character-focused, emotion-focused, or plot-focused mashup videos.
  • Batch production of marketing content: Quickly generate multiple versions of promotional videos from the same batch of footage and different music styles, adapting to a brand's distribution needs on different platforms.
  • Multi-platform short video distribution: Automatically crop and generate multiple aspect ratios such as 9:16 (Douyin/WeChat Channels), 16:9 (Bilibili), and 1:1 (Xiaohongshu), covering all platforms in a single production.
  • Music MV and rhythm-focused content: Use music structure analysis capabilities to strictly align visuals with the musical beat, producing highly rhythmic visual music content or dance videos.