Rott-Ed
Rott-Ed turns study material into short vertical videos: it summarizes a PDF to a 60 second script, picks the idea to illustrate, narrates it and composites everything into a captioned clip. This version runs the same pipeline entirely in your browser, with lightweight stand-ins for the hosted AI models.
- Extract textPDF upload → pdf.js
- 60 second summaryGemini 2.0 Flash → TextRank
- Rank 5 key ideasGemini chat → term weighting
- Cover imageImagen 3 → generative art
- NarrationgTTS → Web Speech API
- Compose clipMoviePy → canvas
Generate a video to see its script.
The real pipeline
The Flask backend sends the PDF to Gemini for a script that reads in 60 seconds, asks it for five visual ideas and to rank them, renders the winner with Imagen, narrates the script with gTTS and composites it all over looping gameplay footage with MoviePy.
What runs here
An extractive TextRank summary sized to the same 60 seconds, ranked key terms, seeded generative art for the cover, your browser's speech engine, and a canvas renderer with word-synced captions where the voice reports word timings.
My part
I built the natural language processing pipeline and worked on the React front end. The platform was tested with a group of more than 20 students.