# FunClip

> Open-source, locally deployed video clipping tool that transcribes speech with FunASR models and cuts clips from selected text or speakers.

FunClip is an open-source video clipping tool from the modelscope organization that uses FunASR Paraformer speech recognition to transcribe videos. Users pick text segments or speakers from the transcript and generate the matching clip, with SRT subtitles returned. The latest listed release is v2.2.1, dated 2026-09-01.

## What It Is

FunClip is a locally deployed, Gradio-based tool for turning a transcribed video into clips. Speech recognition runs on Alibaba's FunASR Paraformer models, which produce sentence-level timestamps. A hosted Hugging Face Space demo offers a simplified Chinese-language flow, and the full app can be installed locally.

## How the Clipping Workflow Works

You upload a video, transcribe it, copy the text segments you want into the clip field, adjust subtitle settings if needed, then click Clip or Clip and Generate Subtitles. The tool supports multi-segment clipping and returns SRT subtitles for the full video and for the target segments. Speaker-based clipping uses the CAM++ speaker recognition model, and hotword customization via SeACo-Paraformer improves recognition of names and entities. A command-line mode offers a recognize stage and a clip stage.

## Models and LLM-Assisted Clipping

The launcher can switch models, including Fun-ASR-Nano, SenseVoice for multilingual ASR with emotion and audio-event tags, and a third-party MOSS-Transcribe-Diarize backend for long-form ASR with anonymous speaker labels. LLM-assisted clipping sends the transcript to a chosen LLM, and the returned segments feed the AI Clip button. Optional gateways include OrcaRouter and Cheaper Inference, and TwelveLabs Pegasus can reason over the video itself rather than only the transcript.

## Setup Path

Installation uses a Python virtual environment, a requirements file and a launch script that serves the UI on localhost:7860. Model weights download separately at startup. Source code is MIT licensed, and model weights follow their own model-page terms.

## Features
- Speech recognition with sentence-level timestamps using Paraformer
- Clip by selected text segments or by speaker
- Multi-segment free clipping
- Automatic SRT subtitles for full video and clipped segments
- Hotword customization with SeACo-Paraformer
- Speaker recognition with CAM++
- LLM-assisted AI clipping
- Optional TwelveLabs Pegasus video-understanding clipping
- Model options: Paraformer, Fun-ASR-Nano, SenseVoice, MOSS
- Local Gradio web UI and command-line mode

## Integrations
FunASR, Gradio, FFmpeg, Hugging Face Spaces, ModelScope, OrcaRouter, Cheaper Inference, TwelveLabs Pegasus, MOSS-Transcribe-Diarize, vLLM

## Platforms
WINDOWS, MACOS, LINUX, WEB, API, CLI

## Pricing
Open Source

## Version
v2.2.1

## Links
- Website: https://huggingface.co/spaces/FunAudioLLM/FunClip
- Repository: https://github.com/modelscope/FunClip
- EveryDev.ai: https://www.everydev.ai/tools/funclip
