FunClip

FunCLIp is an open-source utility for turning long videos into short, targeted clips by working from the spoken content rather than Timeline scrubbing. It runs speech recognition over a video, exposes the transcript, and lets editors pick text passages or specific speakers as the bASIs for a cut. Because the project is Open Source and designed for local use, teams can run it on their own hardware and adapt it to their workflows.

What it does

The tool pAIrs the FunASR Paraformer family of speech reCognition models with a simple browser interface for uploading video, reviewing the transcript, and exporting selected segments. Recognition produces subtitle text, and the user then selects portions of that text or a detected speaker ID to DeFine clip boundaries. An SRT file for the full video and for each chosen segment can be generated alongside the cut footage.

Key capabilities

• Transcript-based clipping: choose text spans in the recognition result and export the matching video portion.
• Speaker-aware cuts: a speaker recognition model assigns speaker IDs so users can clip segments belonging to one person.
• Hotword customization: specific names or domain terms can be supplied as hotwords to improve recognition of those entities.
• Multi-segment editing: several passages can be selected in one pass, with subtitles returned for the full video and the chosen segments.
• large language model assistance: later releases added LLM-guided editing, with integrations such as Qwen and GPT-series models mentioned by the project.
• multilingual support: English audio recognition and clipping were added in mid-2024 alongside Chinese support.
• Local deployment: the interface is built with Gradio and can be launched on a local machine.

Who it is for

FunClip suits podcasters, interview editors, course creators, journalists, and content teams that need to locate and extrACT spoken moments quickly. It is also a reasonable fit for developers and researchers who want an open, inspectable pipeline they can extend with different recognition models or editing rules. Users who prefer a fully managed cloud editor may find the local setup less convenient.

How to get started

Clone the repository, install the Python dependencies, and launch the Gradio service locally, then open the local address in a browser. Upload a video, run recognition, and choose text or speaker targets before exporting clips. Command-line scripts are also provided for staged recognition and clipping, which can help with batch or scripted work. Exact commands and dependency veRSIons may change, so consult the project documentation for current instructions.

Pricing and access caveat

FunClip itself is distributed as open-source software at no cost, but running it requires your own compute and stoRAGe, and any third-party model APIs you connect may cARRy their own fees. Feature availability, model integrations, and hardware requirements can vary by release, so verify the current details before relying on them in production.