Varna is a prototype that automatically transforms research papers into short, engaging videos—like Instagram Reels.
The main idea is to make academic research simple, visual, and easy to understand for everyone.
It extracts key points from research papers, translates them into multiple Indian languages, and creates vertical videos with synchronized voice-over and animated text.
Most research papers remain within academic circles because they are often long, complex, and not presented in a visual or accessible way.
Social media platforms, in contrast, thrive on short and attention-grabbing content.
Varna bridges this gap by turning research findings into 20-second multilingual “research reels” that are easy to understand and share.
Our system uses an automated pipeline that takes a research paper, extracts key points, summarizes and translates them using Bhashini, and creates a natural voice-over using Google TTS.
These elements are then combined with background video and animated text using MoviePy.
The final output is a 1080x1920 vertical video with word-by-word animation and synchronized narration, suitable for Instagram, YouTube Shorts, or educational platforms.
Built with Python, using the following tools and APIs:
- Bhashini API – translations
- Google Text-to-Speech (TTS) – voice narration
- MoviePy – video creation and editing
- Pillow (PIL) – text and image overlays
- dotenv – managing API keys
Varna automatically generates short videos from research papers with animated subtitles, background visuals, and multilingual voice-over.
The system:
- Supports both real API mode (with Bhashini key) and offline mode using mock translations.
- Is modular, making it easy to add new video styles, effects, or templates.
- Fully automates the video generation process, producing a 20-second reel in under a minute.
- Python 3.7+
- FFmpeg installed (required for video rendering)
- Download or clone the project folder:
git clone https://github.com/yourusername/varna.git cd varna - Install dependencies:
pip install -r requirements.txt
- Add your assets:
assets/bg_videos/background.mp4assets/images/robot_img.png
- (Optional) If you have a Bhashini API key, create a
.envfile in the project root and add:The system still runs without it using mock translations.BHASHINI_API_KEY=your_api_key_here
Run the main script:
python main.pyAfter execution, the generated video will appear in:
assets/output/reel_final.mp4
Video Details:
- Resolution: 1080x1920 (vertical)
- Duration: ~20 seconds (5 seconds per research point)
- Languages Supported: English, Hindi, Telugu
To modify the text, edit the list of research points in main.py.
- Extracts 3–4 main research points from the paper.
- Translates them into the selected language using Bhashini.
- Converts translated text into speech using Google TTS.
- Selects a random 20-second segment from the background video.
- Adds word-by-word text animations synced with narration.
- Combines all components into the final vertical reel.
The process is automated and takes less than a minute per video.
- Bhashini API – translates research points into Indian languages
- Google Text-to-Speech (TTS) – generates natural audio narration
- Add a web interface for users to upload papers and choose language/style.
- Introduce background music, multiple templates, and transitions.
- Add analytics to track engagement and optimize video performance.
- Host on the cloud to process videos at scale.
To verify setup:
python test_installation.pyIf all modules pass, a sample video will be generated successfully.
Varna successfully converts text from research papers into short, multilingual reels with synchronized audio and animated captions.
It’s lightweight, social-media ready, and demonstrates how AI translation combined with media automation can make research more accessible.
Varna aims to make academic research more accessible and engaging. By turning dense papers into visual and multilingual reels, it bridges the gap between researchers and learners, enabling quick and inclusive knowledge sharing.