Metadata-Version: 2.4
Name: cutai
Version: 0.1.0
Summary: AI video editor that takes natural language instructions and edits videos locally.
Author-email: Minseo <minseo@example.com>
License-Expression: MIT
Project-URL: Homepage, https://github.com/mindsurf0176-ui/cutai
Project-URL: Repository, https://github.com/mindsurf0176-ui/cutai
Project-URL: Documentation, https://github.com/mindsurf0176-ui/cutai#readme
Project-URL: Issues, https://github.com/mindsurf0176-ui/cutai/issues
Keywords: video,editor,ai,ffmpeg,whisper
Classifier: Development Status :: 3 - Alpha
Classifier: Intended Audience :: Developers
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.10
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Programming Language :: Python :: 3.13
Classifier: Programming Language :: Python :: 3.14
Classifier: Topic :: Multimedia :: Video
Classifier: Topic :: Multimedia :: Video :: Non-Linear Editor
Requires-Python: >=3.10
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: faster-whisper>=1.0
Requires-Dist: scenedetect[opencv]>=0.6
Requires-Dist: typer>=0.9
Requires-Dist: rich>=13.0
Requires-Dist: openai>=1.0
Requires-Dist: pydantic>=2.0
Requires-Dist: pyyaml>=6.0
Provides-Extra: dev
Requires-Dist: pytest>=7.0; extra == "dev"
Requires-Dist: ruff>=0.1; extra == "dev"
Requires-Dist: mypy>=1.0; extra == "dev"
Requires-Dist: pytest-cov>=4.0; extra == "dev"
Dynamic: license-file

# 🎬 CutAI — AI Video Editor with Natural Language

> "Film it. Describe the edit. Done."

[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://github.com/mindsurf0176-ui/cutai/blob/main/LICENSE)
[![Python 3.10+](https://img.shields.io/badge/python-3.10+-blue.svg)](https://www.python.org/downloads/)
[![PyPI version](https://img.shields.io/pypi/v/cutai.svg)](https://pypi.org/project/cutai/)
[![CI](https://github.com/mindsurf0176-ui/cutai/actions/workflows/ci.yml/badge.svg)](https://github.com/mindsurf0176-ui/cutai/actions/workflows/ci.yml)

**CutAI** is an open-source, local-first AI video editor. Give it a video and a sentence — it analyzes scenes, removes silence, adds subtitles, applies color grading, and renders the result. No cloud, no subscription, no upload. Everything runs on your machine.

```bash
$ cutai edit vlog.mp4 -i "remove boring parts, add subtitles, make it warm and cinematic"

🎬 Analyzing vlog.mp4...
  ✅ Detected 23 scenes (12:34)
  ✅ Transcribed 847 words
  ✅ Found 4 silent segments

📋 Edit Plan:
  • Remove 6 low-engagement scenes
  • Remove 4 silent segments
  • Add subtitles (auto-detected: English)
  • Apply warm color grade

🎬 Rendering → output/vlog_edited.mp4
  ✅ Done! 7:21 (trimmed 42%)
```

---

## ✨ Features

### Core Editing
- **Natural language instructions** — describe edits in English or Korean
- **Smart scene detection** — content-aware scene boundary detection via PySceneDetect
- **Silence removal** — automatically finds and cuts dead air
- **Auto-generated subtitles** — powered by Whisper, supports 99+ languages
- **BGM mixing** — add mood-based background music with auto fade-in/out
- **Color grading** — presets: bright, warm, cool, cinematic, vintage
- **Transitions** — fade, dissolve, wipe between scenes
- **Speed control** — speed up, slow down, or set custom playback rates

### 🧬 Edit Style Transfer

Extract the "editing DNA" from any video and apply it to your footage — like LoRA for video editing.

```bash
# Extract how a YouTuber edits
$ cutai style-extract pewdiepie_video.mp4 -o pewdiepie-style.yaml

# Apply that style to your video
$ cutai style-apply myvlog.mp4 --style pewdiepie-style.yaml -o output.mp4

# Learn a channel's style from multiple videos
$ cutai style-learn vid1.mp4 vid2.mp4 vid3.mp4 --name "my-channel" -o channel-style.yaml
```

The Edit DNA captures rhythm (cut pacing, cuts per minute), transitions (fade vs. hard cut ratios), visual tone (brightness, saturation, color temperature), audio mixing (BGM presence, silence tolerance), and subtitle preferences — all as a portable YAML file.

### 🎯 Smart Highlights
- **Engagement scoring** — AI scores each scene (0–100) based on audio energy, speech density, visual activity, and pacing
- **Highlight reels** — auto-generate best-moments, narrative, or shorts-style highlight clips
- **Duration targeting** — specify output length, and CutAI picks the best scenes to fit

### 💬 Interactive Chat Mode
- **REPL-based editing** — iteratively refine edits through conversation
- **Undo/redo** — full history stack, roll back any edit
- **Live preview** — generate quick 360p previews before final render
- **Style loading** — load Edit DNA presets mid-session
- **Slash commands** — `/preview`, `/render`, `/undo`, `/redo`, `/style`, `/plan`, `/help`

### 🔒 100% Local & Free
- **No cloud uploads** — your videos never leave your machine
- **Works offline** — use with Ollama for fully air-gapped operation
- **MIT licensed** — free for personal and commercial use, no usage limits

---

## 🚀 Quick Start

### Prerequisites

```bash
# macOS
brew install ffmpeg

# Ubuntu / Debian
sudo apt-get update && sudo apt-get install -y ffmpeg

# Windows (via Chocolatey)
choco install ffmpeg
```

### Installation

```bash
pip install cutai
cutai --help
```

### Your First Edit

```bash
# Basic edit — remove silence and add subtitles
cutai edit video.mp4 -i "remove silence, add subtitles"

# Apply cinematic look
cutai edit vlog.mp4 -i "자막 추가, 시네마틱하게, 배경음악 넣어줘"

# Interactive chat mode
cutai chat video.mp4

# Quick low-res preview (5–10× faster)
cutai preview video.mp4 -i "remove boring parts"
```

### Style Transfer Workflow

```bash
# 1. Extract style from a reference video
cutai style-extract reference.mp4 -o style.yaml

# 2. View what was extracted
cutai style-show style.yaml

# 3. Apply to your video
cutai edit myvideo.mp4 --style style.yaml

# Or use built-in presets
cutai edit myvideo.mp4 --style cinematic
```

### Smart Highlights

```bash
# Generate a 60-second highlight reel
cutai highlights longvideo.mp4 --duration 60

# Shorts-style: best contiguous segment
cutai highlights longvideo.mp4 --duration 60 --style shorts

# Narrative: keep story arc (hook + ending)
cutai highlights longvideo.mp4 --duration 120 --style narrative

# View engagement scores per scene
cutai engagement video.mp4
```

### Multi-Video Editing

```bash
# Combine clips into one video
cutai multi clip1.mp4 clip2.mp4 clip3.mp4 -i "make a travel montage"

# Combine with style
cutai multi day1.mp4 day2.mp4 -i "remove boring parts" --style vlog-casual
```

---

## 📖 Commands

| Command | Description |
|---------|-------------|
| `cutai edit` | Full pipeline: analyze → plan → render |
| `cutai chat` | Interactive chat-based editing |
| `cutai analyze` | Analyze video (scenes, transcript, quality) |
| `cutai plan` | Generate edit plan without rendering |
| `cutai preview` | Quick low-res preview (360p) |
| `cutai multi` | Combine and edit multiple videos |
| `cutai highlights` | Auto-generate highlight reel |
| `cutai engagement` | Show per-scene engagement scores |
| `cutai style-extract` | Extract Edit DNA from a video |
| `cutai style-apply` | Apply Edit DNA to a video |
| `cutai style-learn` | Learn style from multiple videos |
| `cutai style-show` | Display Edit DNA details |
| `cutai prefs` | View/reset learned preferences |

---

## 🧬 Edit Style Transfer — How It Works

**The Idea:** Every video editor has a style — how fast they cut, which transitions they use, how they grade colors. CutAI extracts these patterns into a structured "Edit DNA" that can be saved, shared, and applied to other videos.

```
┌─────────────┐    extract    ┌───────────┐    apply     ┌──────────────┐
│  Reference   │ ──────────→ │  Edit DNA  │ ──────────→ │  Your Video  │
│  Video       │              │  (YAML)    │              │  (styled)    │
└─────────────┘              └───────────┘              └──────────────┘
```

### Edit DNA Structure

```yaml
name: cinematic
description: Slow pacing, desaturated tones, dramatic transitions
rhythm:
  avg_cut_length: 6.0          # seconds between cuts
  cuts_per_minute: 6.0         # overall pacing
  pacing_curve: slow-fast-slow # how pace changes over time
transitions:
  jump_cut_ratio: 0.5          # 50% hard cuts
  fade_ratio: 0.3              # 30% fades
  dissolve_ratio: 0.15         # 15% dissolves
  avg_transition_duration: 1.0
visual:
  avg_brightness: -0.1         # slightly darker
  avg_saturation: 0.8          # desaturated
  avg_contrast: 1.2            # boosted contrast
  color_temperature: cool
audio:
  has_bgm: true
  bgm_volume_ratio: 0.2
  silence_tolerance: 2.0       # allow longer pauses
subtitle:
  has_subtitles: false
```

Think of it like **image style transfer, but for video editing decisions**. Or like a **LoRA** — a small, portable file that shifts the behavior of the editing pipeline.

### Built-in Presets

| Preset | Pacing | Color | Transitions |
|--------|--------|-------|-------------|
| `cinematic` | Slow (6s avg cuts) | Cool, desaturated | Fades + dissolves |
| `vlog-casual` | Fast (2–3s cuts) | Warm, bright | Jump cuts |

Create your own presets with `cutai style-extract` or write YAML by hand.

---

## ⚙️ Configuration

### Config File

CutAI stores configuration at `~/.cutai/config.yaml`:

```yaml
openai_api_key: ""              # For LLM-powered edit planning
default_whisper_model: base     # tiny/base/small/medium/large
default_llm: gpt-4o            # LLM model for planning
output_dir: ./output            # Default output directory
```

### Environment Variables

| Variable | Description |
|----------|-------------|
| `OPENAI_API_KEY` | OpenAI API key (enables LLM-based planning) |
| `CUTAI_WHISPER_MODEL` | Default Whisper model size |
| `CUTAI_LLM` | Default LLM model |
| `CUTAI_OUTPUT_DIR` | Default output directory |
| `CUTAI_FFMPEG_PATH` | Custom FFmpeg binary path |

### Local-Only Mode (No API Key)

CutAI works without any API key using **rule-based planning**. Common instructions like "remove silence", "add subtitles", "make it cinematic" are handled by built-in rules — no LLM needed.

For more complex instructions, set `OPENAI_API_KEY` or use a local LLM via Ollama.

### Personal Learning

CutAI learns from your editing patterns over time. Preferences are stored locally at `~/.cutai/learning.json` — your editing history never leaves your machine.

```bash
# View learned preferences
cutai prefs

# Reset learning data
cutai prefs --reset
```

---

## 🏗️ Architecture

```
┌─────────────────────────────────────────────────────────────┐
│                        CutAI Pipeline                       │
│                                                             │
│  ┌──────────┐  ┌──────────┐  ┌──────────┐  ┌────────────┐  │
│  │ Analyzer │→│ Planner  │→│  Editor  │→│  Renderer  │  │
│  │          │  │          │  │          │  │            │  │
│  │• Scenes  │  │• Rules   │  │• Cutter  │  │• FFmpeg    │  │
│  │• Whisper │  │• LLM     │  │• Subtitle│  │• Concat    │  │
│  │• Quality │  │• Style   │  │• BGM     │  │• Filters   │  │
│  │• Energy  │  │• Learning│  │• Color   │  │            │  │
│  │          │  │          │  │• Speed   │  │            │  │
│  │          │  │          │  │• Trans.  │  │            │  │
│  └──────────┘  └──────────┘  └──────────┘  └────────────┘  │
│                                                             │
│  ┌──────────────────────────────────────────────────────┐   │
│  │ Style Engine: Extractor → DNA (YAML) → Applier      │   │
│  └──────────────────────────────────────────────────────┘   │
│  ┌──────────────────────────────────────────────────────┐   │
│  │ Engagement Engine: Audio + Visual + Structural → 0-100│  │
│  └──────────────────────────────────────────────────────┘   │
└─────────────────────────────────────────────────────────────┘
```

### Module Breakdown

```
cutai/
├── cli.py                  # Typer CLI with Rich formatting
├── config.py               # Config management (~/.cutai/)
├── chat.py                 # Interactive REPL chat mode
├── highlight.py            # Smart highlight generation
├── learning.py             # Personal preference learning
├── multi.py                # Multi-video editing pipeline
├── preview.py              # Low-res preview renderer
├── models/
│   └── types.py            # Pydantic v2 models (30+ types)
├── analyzer/
│   ├── scene_detector.py   # PySceneDetect integration
│   ├── transcriber.py      # Whisper transcription
│   ├── quality_analyzer.py # Silence detection, audio analysis
│   └── engagement.py       # Engagement scoring engine
├── editor/
│   ├── cutter.py           # Cut/trim operations
│   ├── subtitle.py         # ASS subtitle generation
│   ├── bgm.py              # Background music mixing
│   ├── color.py            # Color grading (FFmpeg filters)
│   ├── speed.py            # Speed adjustment
│   ├── transition.py       # Scene transitions
│   └── renderer.py         # Final render orchestration
├── planner/
│   └── edit_planner.py     # Rule-based + LLM planning
└── style/
    ├── extractor.py        # Extract Edit DNA from video
    ├── applier.py          # Apply Edit DNA to video
    ├── learner.py          # Learn style from multiple videos
    ├── io.py               # YAML serialization
    └── presets/            # Built-in style presets
        ├── cinematic.yaml
        └── vlog-casual.yaml
```

---

## 🤝 Contributing

We welcome contributions! See [CONTRIBUTING.md](CONTRIBUTING.md) for the full guide.

### Quick Start

```bash
git clone https://github.com/mindsurf0176-ui/cutai.git
cd cutai
python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
pytest tests/ -v
```

### Ways to Contribute

- 🐛 **Bug reports** — open an issue with reproduction steps
- 🧬 **Style presets** — create and share Edit DNA presets
- 🔧 **New edit operations** — add operations in `cutai/editor/`
- 📖 **Documentation** — improve docs, add examples
- 🌍 **Translations** — help with i18n for rule-based patterns

---

## 📊 Benchmarks

Processing time depends on video length, Whisper model size, and whether LLM planning is used.

| Video Length | Analyze | Plan (rules) | Plan (LLM) | Render | Total (rules) |
|-------------|---------|---------------|-------------|--------|---------------|
| 1 min | ~5s | <1s | ~3s | ~10s | ~15s |
| 5 min | ~15s | <1s | ~5s | ~45s | ~60s |
| 15 min | ~40s | <1s | ~8s | ~2min | ~3min |
| 60 min | ~3min | <1s | ~15s | ~8min | ~11min |

*Measured on M1 MacBook Pro, Whisper `base` model, 1080p source.*

### Hardware Requirements

- **Minimum:** 4GB RAM, any modern CPU, FFmpeg installed
- **Recommended:** 8GB+ RAM, Apple Silicon or modern x86-64
- **Whisper models:** `tiny` (39MB) → `large` (3GB) — bigger = more accurate but slower

---

## 🗺️ Roadmap

- [x] Core editing (cuts, subtitles, BGM, color grading, transitions, speed)
- [x] Edit Style Transfer (DNA extraction + application + learning)
- [x] Engagement scoring + smart highlights
- [x] Interactive chat mode with undo/redo + preview
- [x] Multi-video editing
- [x] Personal learning from editing patterns
- [x] Built-in style presets
- [ ] GPU acceleration (CUDA / Metal)
- [ ] Web UI
- [ ] Plugin system for custom operations
- [ ] Community style preset marketplace
- [ ] Real-time preview during chat
- [ ] Batch processing mode

---

## License

[MIT](LICENSE) — free for personal and commercial use.

---

## Star History

[![Star History Chart](https://api.star-history.com/svg?repos=mindsurf0176-ui/cutai&type=Date)](https://star-history.com/#mindsurf0176-ui/cutai&Date)
