notes-to-video — плагин для Claude Code: видео из конспектов

# notes-to-video

[![GitHub stars](https://img.shields.io/github/stars/cymcymcymcym/notes-to-video.svg?style=social)](https://github.com/cymcymcymcym/notes-to-video)
[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
[![Python 3.10+](https://img.shields.io/badge/python-3.10+-blue.svg)](https://www.python.org/downloads/)
[![npm](https://img.shields.io/npm/v/notes-to-video.svg)](https://www.npmjs.com/package/notes-to-video)
[![Platform](https://img.shields.io/badge/platform-linux%20%7C%20macos%20%7C%20windows-lightgrey.svg)]()
[![Claude Code](https://img.shields.io/badge/Claude%20Code-skill-blueviolet.svg)](https://claude.com/claude-code)

Turn notes (LaTeX, PDF, or plain text) into animated explainer videos in the style popularized by 3Blue1Brown — using Manim, TTS, and ffmpeg.

<p align="center">
  <img src="docs/teaser.gif" alt="Change your notes into a 3Blue1Brown-style video" width="720">
</p>

A [Claude Code](https://claude.com/claude-code) skill that handles the full pipeline: content extraction, narration writing with cue markers, Manim scene generation with audio-video sync, validation, rendering, and composition.

## Demo

Feed it lecture notes, get an animated explainer video. **🔊 Turn sound on** — the video has narration.

<table>
<tr>
<td width="40%" align="center"><strong>Input: CS229 Lecture Notes</strong></td>
<td width="60%" align="center"><strong>Output: Animated Explainer</strong></td>
</tr>
<tr>
<td>

<picture>
  <source media="(prefers-color-scheme: dark)" srcset="docs/cs229_vae_notes_dark.png">
  <img src="docs/cs229_vae_notes.png" alt="CS229 VAE lecture notes" width="100%">
</picture>

*Section 11.5 — Variational Auto-Encoder*
*[CS229 Lecture Notes](https://cs229.stanford.edu/main_notes.pdf) — Andrew Ng & Tengyu Ma, Stanford University*
*Used as demo input with attribution.*

</td>
<td>

<video src="https://github.com/user-attachments/assets/c2c9d8b3-cdeb-429d-8b9d-ec9a97bf57a1" poster="docs/vae_thumbnail.png" controls style="max-height:400px; width:100%;"></video>

*🔊 Sound on. 3 min video, generated from notes in one command. [Download](https://github.com/cymcymcymcym/notes-to-video/releases/download/v1.0.0/vae_explainer_captioned.mp4)*

</td>
</tr>
</table>

## Install

This is a Claude Code skill, not a standalone CLI. The npm package is a one-shot installer that drops the skill files where Claude Code can find them.

**1. Install the skill files:**

```bash
npx notes-to-video
```

This copies:
- `skills/notes-to-video/` → `~/.claude/skills/notes-to-video/`
- `video_utils/` → `~/tools/video_utils/` (shared Python helpers the skill imports)

Re-run any time to upgrade.

**2. Set up the Python environment** (one-time, shared across projects):

```bash
python3 -m venv ~/tools/.venv
~/tools/.venv/bin/pip install manim edge-tts pydub faster-whisper
# Optional: chatterbox-tts (local voice cloning), torch with CUDA
```

**3. (Optional) Add API keys** for cloud TTS backends:

```bash
mkdir -p ~/tools/credentials
cat > ~/tools/credentials/.env <<'EOF'
MINIMAX_API_KEY=...
MINIMAX_GROUP_ID=...
OPENAI_API_KEY=...
EOF
```

### Alternative install methods

**Claude Code plugin:**
```bash
/plugin marketplace add cymcymcymcym/notes-to-video
/plugin install notes-to-video@notes-to-video-marketplace
```

**Manual:** clone this repo, copy `skills/notes-to-video/` to `~/.claude/skills/` and `video_utils/` to `~/tools/video_utils/`.

## Quick Start

1. Open Claude Code in your project and ask:
   ```
   make a 3b1b-style video from my_notes.tex
   ```

2. Claude will:
   - Extract key concepts from your notes
   - Write a narration script with cue markers
   - Generate Manim scenes synced to the narration
   - Validate all scenes for visual issues
   - Hand you the build command

## Features

- **Notes to video pipeline** — feed in LaTeX, PDF, or plain text notes, get animated explainer videos
- **Audio-video sync** — cue-based system that synchronizes Manim animations to narration timestamps
- **CText kerning fix** — workaround for Manim's broken Pango kerning ([manim #2844](https://github.com/ManimCommunity/manim/issues/2844))
- **4 TTS backends** — Edge-TTS (free, default), MiniMax (best quality), Chatterbox (local + voice cloning), OpenAI
- **Scene validator** — catches text overlaps, out-of-bounds elements, text overflow, and line-through-text issues before rendering
- **Cross-platform** — Linux, macOS, Windows

## TTS Options

| Backend | Quality | Cost | Requirements |
|---------|---------|------|-------------|
| **Edge-TTS** (default) | Good | Free | None |
| **MiniMax** | Best | ~$0.04/min | API key |
| **Chatterbox** | Good + voice cloning | Free | NVIDIA GPU |
| **OpenAI TTS** | Good | ~$0.06/min | API key |

## How It Works

The core innovation is the **cue-based audio-video sync system**:

1. Narration is written with `{CUE_NAME}` markers at visual event points
2. TTS generates per-sentence audio and estimates cue positions by character ratio
3. Manim scenes read cue timestamps and sync animations accordingly
4. `until()` fills gaps with slow animations, `sync()` waits for exact cue times

This produces smooth, naturally-paced videos where animations fire exactly when the narrator says the relevant keyword.

## Project Structure

Each video is a self-contained `<project>/` subfolder. Adding a second video is zero migration — just create another.

```
final/                               # THE DELIVERABLE — what you watch/share
  <project>/
    <project>.pdf                    # source paper, if applicable
    <project>.mp4                    # final video
    <project>.srt                    # soft subtitles (sidecar)
    <project>_captioned.mp4          # optional: burned-in captions

intermediate/                        # everything else (heavy; .gitignore by default)
  <project>/
    src/
      video_<project>.py             # Manim scenes
      part_<project>_narration.py    # narration with {CUE} markers
      generate_tts_<project>.py      # TTS runner
      build_<project>.py             # render + mux + caption
      assets/<project>/*.png         # extracted source figures
    audio/video_<project>/           # TTS output + durations.json
    media/videos/video_<project>/    # manim render cache
    review/video_<project>/          # validator screenshots
    output/                          # per-scene muxed MP4s
    plan_<project>.md                # scene-by-scene plan
```

The skill ships with:

```
video_utils/                # Bundled library (→ ~/tools/video_utils/ on install)
  manim_helpers.py          # CText, colors, sync helpers
  tts_edge.py              # Edge-TTS (free, default)
  tts_minimax.py           # MiniMax TTS (cloud)
  tts_local.py             # Chatterbox + Whisper (local)
  tts_openai.py            # OpenAI TTS (cloud)
  validate_scenes.py       # Scene validator
  captions.py              # SRT generator

skills/notes-to-video/
  SKILL.md                 # Claude Code skill definition
```

## Acknowledgments

- **[Manim Community](https://www.manim.community/)** — the animation engine this project is built on. Manim was originally created by [Grant Sanderson](https://www.3blue1brown.com/) (3Blue1Brown) and is now maintained by the Manim Community (MIT License).
- **[3Blue1Brown](https://www.3blue1brown.com/)** — the visual style this project emulates. Grant Sanderson's videos set the standard for mathematical explainer animation.
- **[CS229 Lecture Notes](https://cs229.stanford.edu/main_notes.pdf)** — Andrew Ng & Tengyu Ma, Stanford University. Used as source material for the VAE demo video (Section 11.5), with attribution.

**Disclaimer:** This project is not affiliated with, endorsed by, or sponsored by 3Blue1Brown, Grant Sanderson, Stanford University, or the Manim Community. "3Blue1Brown" is used here as a descriptive reference to the visual style.

## License

MIT — see [LICENSE](LICENSE) for details.