PreenCut is an intelligent video editing tool that automatically analyzes audio/video content using speech recognition and large language models. It helps you quickly find and extract relevant segments from your media files using natural language queries.

• Automatic Speech Recognition: Powered by WhisperX for accurate transcription
• AI-Powered Analysis: Uses large language models to segment and summarize content
• Natural Language Querying: Find clips using descriptive prompts like "Find all product demo segments"
• Smart Clipping: Select and export segments as individual files or merged video
• SRT Export: Generate subtitles with custom line length
• Batch Processing: find a specific topic across multiple files
• Re-analysis: Experiment with different prompts without reprocessing audio

1. Clone the repository:

git clone https://github.com/roothch/PreenCut.git
cd PreenCut

1. Install dependencies, recommend using Python 3.11:

pip install -r requirements.txt

1. Install FFmpeg (required for video processing):

# ubuntu/Debian
sudo apt install ffmpeg

# CentOS/RHEL
sudo yum install ffmpeg

# macOS (using Homebrew)
brew install ffmpeg

# Windows: Download from https://ffmpeg.org/

1. Set up API keys (for LLM services): First you need to set your llm services in LLM_MODEL_OPTIONS of config.py. Then set your API keys as environment variables:

# for example, if you are using DeepSeek and DouBao as LLM services
export DEEPSEEK_V3_API_KEY=your_deepseek_api_key
export DOUBAO_1_5_PRO_API_KEY=your_doubao_api_key

1. (Optional)set up gradio temp file directory: set os.environ['GRADIO_TEMP_DIR'] in config.py file.

1. Start the Gradio interface:

python main.py

1. Access the web interface at http://localhost:7860
2. Upload video/audio files (supported formats: mp4, avi, mov, mkv, ts, mxf, mp3, wav, flac)
3. Configure options:

• Select LLM model
• Choose Whisper model size (tiny → large-v3)
• Add custom analysis prompt (Optional)

1. Click "Start Processing" to analyze content
2. View results in the analysis table:

• Start/end timestamps
• Duration
• Content summary
• AI-generated tags

1. Use the "Re-analyze" tab to experiment with different prompts
2. Use the "Cut" tab to select segments and choose export mode:

• Export as ZIP package
• Merge into a single video file

1. you can also visit the Restful api use the route prefix /api/xxx

   • upload file

     POST /api/upload

     body: formdata

     key	value type
     file	file

     reponse: json

       { file_path: f'${GRADIO_TEMP_DIR}/files/2025/05/06/uuid.v1().replace('-', '')${file_extension}' }
     

   • create task

     POST /api/tasks

     body: json

     {
       "file_path": "put the file path here response from upload api, starting with ${GRADIO_TEMP_DIR}",   
       "llm_model": "DeepSeek-V3-0324",
       "whisper_model_size": "large-v2",
       "prompt": "提取重要信息，时间控制在10s"
     }

     response:

       { "task_id": "" }

   • query task reult

     GET /api/tasks/{task_id}

     response:

     {
       "status": "completed",
       "files": [
           "${GRADIO_TEMP_DIR}/files/2025/06/23/608ecc80500e11f0b08a02420134443f.wav"
       ],
       "prompt": "提取重要信息，时间控制在10s",
       "model_size": "large-v2",
       "llm_model": "DeepSeek-V3-0324",
       "timestamp": 1750668370.6088192,
       "status_info": "共1个文件，正在处理第1个文件",
       "result": [
           {
               "filename": "608ecc80500e11f0b08a02420134443f.wav",
               "align_result": {
                   "segments": [
                       {
                           "text": "有内地媒体报道,嫦娥6号着陆器上升器组合体已经完成了钻取采样,接着正按计划进行月面的表取采样。",
                           "start": 1.145,
                           "end": 9.329
                       }
                   ],
                   "language": "zh"
               },
               "segments": [
                   {
                       "start": 1.145,
                       "end": 9.329,
                       "summary": "嫦娥6号着陆器上升器组合体已完成钻取采样，正进行月面表取采样。",
                       "tags": [
                           "嫦娥6号",
                           "月球采样",
                           "航天科技"
                       ]
                   }
               ],
               "filepath": "${GRADIO_TEMP_DIR}/files/2025/06/23/608ecc80500e11f0b08a02420134443f.wav"
           }
       ],
       "last_accessed": 1750668836.8038888
     }

python3 -m uvicorn main:app --port 7860 --reload

• Use faster-whisper for shorter segments, use whisperx for faster processing
• Adjust WHISPER_BATCH_SIZE based on available VRAM
• Use smaller model sizes for CPU-only systems
• If you don't need srt files, you could set ENABLE_ALIGNMENT=False to improve processing speed.

This project is licensed under the MIT License. See the LICENSE file for details.

• Email: 1242727205@qq.com
• RedNote(小红书): 一去二三里
• RedNote Group