LibreAI / Clips courts / FunClip
// clipping · modelscope/FunClip

FunClip

FunClip est un projet de clips courts open source, créé par modelscope (modelscope/FunClip sur GitHub). Il compte 6 379 étoiles sur GitHub et 756 forks, il est publié sous la licence MIT et a été mis à jour pour la dernière fois sur GitHub le 8 octobre 2026.

FunASR-powered video transcription, subtitles and LLM-assisted clipping with a local Gradio UI.

★ 6,4 k étoiles⑂ 756 forksMITmis à jour oct. 2026
▦ visuel — le README de ce projet ne contient aucune capture intégrable,
voici donc un emplacement réservé. Le guide ci-dessous est complet malgré tout.
// détails du projet

Détails du projet

Étoiles GitHub
6 379
Forks
756
Licence
MIT
Dernière mise à jour sur GitHub
8 octobre 2026
Catégorie
Clips courts
Type de fiche
Guide d'installation complet
Site officiel
Official site
// fonctionnement

Comment ça marche

  1. Clarify your clipping goal

    Gather the videos you want to transcribe or cut, and decide whether you need subtitles, speaker-specific clips or highlight reels.

  2. Clone and set up

    Clone the repo, create the documented Python 3.12 virtual environment and install the requirements.

  3. Configure the ASR model

    Pick a model — the default Paraformer pipeline, Fun-ASR-Nano, SenseVoice or the MOSS backend — and let the weights download.

  4. Clip your first video

    Launch the Gradio UI, upload a video, copy the transcript segments you want, and export the clip with subtitles.

// fonctionnalités

Pourquoi on l'adore

Industrial-grade ASR

Integrates Alibaba's Paraformer-Large, one of the best-performing open-source Chinese ASR models, plus Fun-ASR-Nano and SenseVoice options.

Hotword customization

Uses SeACo-Paraformer hotwords so names and domain terms you specify are recognized more accurately.

Speaker-aware clipping

The CAM++ speaker recognition model lets you clip the segments spoken by a specific speaker.

Text-based clipping workflow

Select the transcript text you want and FunClip cuts those segments, returning full-video SRT and target-segment SRT subtitles.

Local Gradio interface

Everything runs through a browser UI that you can host on your own server; no video leaves your machine unless you choose a cloud LLM gateway.

LLM-assisted clipping

Optional LLM integration helps pick and refine clip segments from the transcript.

Advertisement
// lancez-le

Lancez-le vous-même

  1. Clone the repository

    Clone FunClip and create the documented Python environment.

    git clone https://github.com/modelscope/FunClip.git
    cd FunClip
    python3.12 -m venv .venv && . .venv/bin/activate
  2. Install dependencies

    Install the Python requirements and verify the environment.

    python -m pip install -r requirements.txt
    python -m pip check
  3. Launch the Gradio UI

    Start FunClip locally and open the interface in your browser.

    python funclip/launch.py
// cas d'usage

Ce que vous pouvez créer

Advertisement
// faq

Questions fréquentes

Who maintains FunClip?

FunClip is a modelscope project, built on Alibaba's FunASR and Paraformer model family.

Is FunClip free?

The code is MIT licensed and free; the ASR model weights are downloaded separately and governed by the terms on their model pages.

Does it need a paid API key?

Core transcription and clipping run locally with downloaded open models — no key needed. Optional LLM gateways for assisted clipping use your own provider key.

How does it compare to Whisper-based clippers?

FunClip is built around Paraformer/FunASR, which is especially strong for Chinese speech, and adds hotwords and speaker-based clipping in one Gradio tool rather than a bare transcription script.

How do I get started?

Clone the repo, create the venv, install requirements, run python funclip/launch.py and upload a video.

// continuez l'exploration

Outils similaires

Clips courts

ViralMint

Local-first video pipeline: trend scouting, clipping, captioned shorts.

★ 122 étoilesAGPL-3.0
Voir le guide →
Clips courts

velocut

AI-native, local-first video editor by Ribbi — runs entirely in the browser. Rust/WASM engine, WebGPU compositing, WebCodecs export; humans and LLM agents ed…

★ 455 étoilesMIT
Voir le guide →
Clips courts

ai-highlight-clip

Desktop highlight editor with Whisper transcription and LLM clip scoring.

★ 105 étoilesMIT
Voir le guide →
Clips courts

autoclip

Local-first clipper: Whisper plus Ollama turn long video into captioned 9:16 clips.

★ 162 étoilesMIT
Voir le guide →
Clips courts

agentic-video-editor

AI-powered video editor that turns raw footage and a creative brief into a polished ad using an ensemble of AI agents (Google Gemini + FFmpeg)

★ 495 étoilesMIT
Voir le guide →
Clips courts

Gausian_native_editor

Gausian - Rust-based local video editor for AI video production

★ 1,0 k étoilesUnknown
Voir le guide →
// alternatives

Alternatives en Clips courts

Les autres projets de la catégorie Clips courts avec le plus d'étoiles du catalogue LibreAI — la sélection réaliste d'alternatives si FunClip ne vous convient pas.

// plus en clipping

Plus en Clips courts

Projets de la catégorie Clips courts proches dans le catalogue, les plus proches en étoiles de FunClip — continuez à explorer de part et d'autre.

Obtenir FunClip

Gratuit et ouvert : récupérez le code source, mettez une étoile au dépôt et commencez à créer.