English | 简体中文

video-to-notes

video-to-notes is a Codex skill and helper toolkit for turning lecture source material into professional Chinese LaTeX course notes and a compiled PDF.

The main workflow is:

transcript TXT + slide PDF/PPT/PPTX -> slide images -> LaTeX notes -> rendered PDF

If a YouTube URL and cookies.txt are provided, the workflow can first download subtitles only, convert them to TXT, and then use the transcript with the local slides.

Repository Layout

• SKILL.md: the Codex skill instructions. This is the source of truth for the writing, figure-selection, subtitle-download, LaTeX, and validation workflow.
• assets/notes-template.tex: the default Chinese LaTeX note template, including a simple title page, highlight boxes, listings, figures, and TikZ.
• scripts/prepare_slide_images.py: renders a slide PDF into pic/page_N.png images. It can auto-discover one PDF and one TXT in the current directory.
• scripts/srt_to_txt.py: converts .srt or .vtt subtitles into a readable .txt transcript.
• clean.sh: archives the current lecture workspace into a named folder after notes are generated.
• agents/openai.yaml: a short agent-facing entry point for invoking the skill.
• demo/: sample slide PDFs for testing the image-preparation helper.

Supported Inputs

• Local transcript TXT plus local slide PDF.
• Local transcript TXT plus local slide PPT/PPTX, after converting the deck to PDF.
• YouTube URL or url.txt plus cookies.txt, then downloaded subtitles plus local slides.

When there are multiple candidate transcripts or slide decks, choose the files explicitly instead of relying on auto-discovery.

Prerequisites

Python helpers use only the standard library except for optional PDF rendering support:

python3 -m pip install pdf2image

Install Poppler so PDF pages can be rendered. On macOS:

brew install poppler

On Ubuntu/Debian:

sudo apt-get install poppler-utils

For optional YouTube subtitle download, install yt-dlp:

python3 -m pip install yt-dlp

For final PDF compilation, install a LaTeX distribution with XeLaTeX and common packages such as ctex, tcolorbox, listings, tikz, and pgfplots.

Quick Start

1. Prepare Source Files

Place the lecture transcript and slide deck in the working directory. If the transcript is in subtitle format, convert it first:

python3 scripts/srt_to_txt.py lecture.srt transcript.txt

2. Render Slide Images

If the working directory contains exactly one .pdf and one .txt, use auto-discovery:

python3 scripts/prepare_slide_images.py --auto

Otherwise, pass the slide PDF explicitly:

python3 scripts/prepare_slide_images.py --pdf slides.pdf --output pic

The helper writes images as pic/page_1.png, pic/page_2.png, and so on. If pic/ already contains page images, it reuses them by default. Add --force to re-render.

To test with a demo deck:

python3 scripts/prepare_slide_images.py --pdf demo/karpathy_llm_intro.pdf --output pic --force

3. Generate the Notes

Ask Codex to use the video-to-notes skill with your transcript and slides. The skill will:

1. inspect the sources and slide images,
2. use assets/notes-template.tex as the base document,
3. select or create teaching figures,
4. write a complete Chinese note,
5. compile the .tex file to PDF,
6. validate that referenced assets exist and the PDF builds.

For manual use, start from assets/notes-template.tex, replace the body block, and reference slide screenshots with the pic/page_N.png convention.

4. Compile Manually When Needed

xelatex notes.tex
xelatex notes.tex

Run XeLaTeX twice when the table of contents or references need a second pass.

5. Archive a Finished Lecture

chmod +x clean.sh
./clean.sh Lec_01_Intro

clean.sh creates the target folder and moves pic/, all root-level *.pdf, *.txt, and *.tex files into it. It does not currently move .srt or .vtt files.

Output Conventions

• Notes are written in Chinese unless requested otherwise.
• Slide screenshots should use pic/page_N.png; captions should explain the teaching role without adding source page numbers.
• Figures should stay outside importantbox, knowledgebox, warningbox, and dialoguebox.
• Each major section should end with 本章小结.
• The document should end with a top-level 总结与延伸 section.
• Final delivery should include the generated .tex, compiled .pdf, and every referenced asset.