Revisor

AI-powered video transcription — 100% local and free.

Revisor is a web application that transcribes video files and YouTube links into text using OpenAI's Whisper turbo model. All processing happens on the user's machine, with no data sent to external services (except Brevo for email, when configured).

---

Features

• Video upload — drag-and-drop support for MP4, AVI, MOV, MKV, WebM and other formats (up to 2GB)
• YouTube — support for individual URLs and playlists (up to 20 videos)
• Smart queue — add multiple videos at once; they are processed one by one in the background
• AI correction — uses local Ollama to fix grammar and format paragraphs
• Promotion suggestions — automatically generates promotional text
• Clip suggestions — identifies highlight moments for short-form content
• Video clipping — cut segments with quality control (CRF) and optional watermark removal
• Email delivery — transcription sent as a ZIP via Brevo, manually or automatically
• Completion notification — email alert when the entire queue is done
• Download — transcription in TXT, segments in SRT and CSV

---

Prerequisites

Tool	Version
Python	3.10+
pip	22+
ffmpeg	4.4+
Ollama	Optional (for AI correction)

---

Installation

1. System dependencies

sudo apt update && sudo apt install -y ffmpeg

2. Python environment

python3 -m venv venv
source venv/bin/activate
pip3 install -r requirements.txt

3. (Optional) Configure email

Create a .env file in the project root:

BREVO_API_KEY=your_brevo_api_key
BREVO_MAIL_NAME=Revisor
BREVO_MAIL_SENDER=mail@mence.dev

4. (Optional) Install Ollama

curl -fsSL https://ollama.ai/install.sh | sh
ollama pull qwen2.5:7b-instruct-q3_K_M

---

Usage

source venv/bin/activate
python3 server.py

Open http://localhost:5000 in your browser.

Basic workflow

1. On the Upload page (/transcrever), add files or paste YouTube links
2. Set the language and optionally provide an email to receive the transcription
3. Check "Notify when finished" to get an alert when the queue is complete
4. Go to Files (/arquivo) and click Process
5. Follow the progress in real time
6. Click a completed item to view the transcription, suggestions, and clips

---

Pages

Route	Page	Purpose
/	Home	Project info, features, statistics
/transcrever	Upload	Add files and YouTube URLs
/arquivo	Process & Files	Processing queue, results, video clipping

---

Project structure

revisor/
├── server.py              # HTTP server (Flask) + queue + email
├── transcriber.py         # Transcription engine (Whisper + Ollama)
├── test_app.py            # Test suite (28 tests)
├── templates/
│   ├── index.html         # Home page
│   ├── transcrever.html   # Video upload
│   └── arquivo.html       # Queue, results and clips
├── static/
│   └── styles.css         # Dark theme (~4700 lines)
├── docs/                  # Architecture documentation
└── uploads/               # Temporary files

---

Known limitations

• CPU: Whisper runs in FP32 without an NVIDIA GPU with CUDA — significantly slower
• Sequential processing: videos are processed one at a time
• Memory: very long videos (>1h) may require substantial RAM
• Ollama: AI correction features are optional; if Ollama is not available, they are skipped

---

Support

Built by Mence — Support on Ko-fi

---

License

100% free and open source. Educational, personal, and commercial use allowed.