01
Clean, occlusion-free slides
Real-pixel compositing rebuilds the parts of a slide the presenter, cursor, or laser was covering — reconstructed from other frames, never invented by a model.
Graphics and audio, not transcription alone
Turn any lecture into clean slides and notes.
A recording is not a study resource. PairedY pulls every slide out of the video, cleans it, reads it with a vision model, and pairs it back to what the speaker actually said.
Rebuilt from real pixels — occlusions filled from other frames, never hallucinated.
Every claim keeps its timestamp — jump from any note back to the second it came from.
Yours to keep — PPTX, PDF, Markdown, and a portable knowledge bundle.
26slides recovered
1,326transcript segments aligned
24 minlecture, start to finish
10output files
A real run: one 24-minute ML lecture, converted end to end. Slides, an annotated handout, an editable deck, notes, and a knowledge graph.
- No login
- Runs on our workers
- Bring your own model
- Open-source
- Exports you own
What makes it different
Faithful. Aligned. Yours.
02
Decks and whiteboards
Handles projected slides and chalk or marker boards alike — illumination normalization, stroke boosting, erase-aware segmentation. Detected automatically.
03
Reads the board and the room
Per-slide OCR of text, equations, code, tables, and figures, aligned against the timestamped narration so each slide carries what was actually said about it.
04
Decks you can actually edit
An image-backed PPTX with speaker notes, an editable reconstruction, and an annotated handout with the narration set in the margin.
05
A knowledge bundle, not a transcript
One provenance-rich export: typed nodes and edges, timestamps, deep links back into the video, and chunks at several granularities. RAG and GraphRAG ready.
06
Your model, your key
Any OpenAI-compatible endpoint or OpenRouter for the vision and language steps. Ask AI runs from your browser with your own key — it never reaches our servers.
How it works
Paste a link. Come back to a deck.
Paste a link. A YouTube URL, or upload a file you already have.
We find the slides. Boundary detection, crop and deskew, de-occlusion, dedupe.
We read and align. Vision-model OCR fused with the timestamped transcript.
You get the outputs. Slides, PPTX, an annotated handout, notes, and the knowledge bundle.
The lecture you can study.
Conversion runs on our workers — no key needed to start. Bring your own model when you want to ask questions of the result.