jianjieyiban/JJYB_AI_VideoAutoCut

JJYB AI VideoAutoCut – One‑stop AI video editing desktop app

What it is – A Windows‑only desktop application (Flask + PyWebView) that lets a creator take raw video clips, have the software analyse them, automatically generate a script, synthesize a voice‑over, sync everything on a timeline and export the final cut. All data lives locally (SQLite, FFmpeg, OpenCV) and the UI is bilingual (Chinese / English).

Core capabilities

Feature	What you get
Smart shot segmentation	Uses TransNetV2 for deep shot detection; falls back to OpenCV frame‑difference when the model isn’t available.
AI script generation	Calls a configurable LLM/VLM to produce a narration script in 12 preset styles (e.g., news, vlog, documentary) plus fine‑grained control over tone, hook, persona, speed and emotion.
Multi‑engine TTS & voice cloning	Supports Edge, Azure, gTTS, Coqui, etc., and local IndexTTS2 voice‑cloning (requires a separate Python 3.10 environment).
Puppet‑sync engine	Aligns each script segment with the best matching shot, handling cuts, slow‑motion and high‑light highlights without fabricating frames.
Original audio mixing	Optionally inserts portions of the source audio at configurable ratios.
Target‑duration enforcement	Guarantees the final video matches a user‑defined length, verified with FFprobe.
AI‑assisted remix	Two remix modes – normal (shot‑scoring) and music‑beat (beat detection via librosa) – with preset rhythm packages for trailers, game highlights, travel montages, etc.
AI cover generation	Calls cloud image models (DALL‑E 3, Midjourney‑like services) or a local fallback template to create thumbnail images.
Full‑featured timeline editor	Multi‑track video/audio/subtitle editing, basic filters, transitions and playback speed control.
Export to Jianying (CapCut) draft	Generates a native CapCut project folder so the user can continue editing in the mobile app.
Project & asset management	Local SQLite stores projects, material library, and logs; drag‑and‑drop upload, tagging, search, batch actions.
Bilingual UI	All UI strings (200+) are translated; language choice persists across sessions.

Tech stack (high‑level)

• Frontend: Flask + Flask‑SocketIO, HTML/JS (Layui), PyWebView for a native window.
• Video/audio: FFmpeg/ffprobe, OpenCV, MoviePy, librosa, Pydub.
• AI models: Configurable cloud LLM/VLM (OpenAI, Claude, Gemini, etc.), optional local PyTorch models (TransNetV2, YOLO, Whisper), IndexTTS2 for voice cloning.
• Data: SQLite, local file system (uploads/, output/, temp/).
• OS: Windows 10/11, optional CUDA for faster deep‑learning inference.

Getting started (quick summary)

1. Download the latest release zip.
2. Install Python 3.10‑3.13 and run pip install -r requirements.txt.
3. Install FFmpeg and add it to PATH.
4. Double‑click 启动应用.bat (or run python frontend/app.py).
5. Open the web UI (usually http://127.0.0.1:5000).
6. In Settings add at least one LLM API key, a vision model if you want automatic shot analysis, and an image‑generation API if you need AI covers.

Typical workflow

1. Import raw footage via the Material Library.
2. Smart segmentation → list of shots.
3. Generate / edit script using the chosen narration style.
4. Generate voice‑over (choose TTS or cloned voice).
5. Puppet‑sync aligns shots with voice, optionally mixes original audio.
6. Preview / fine‑tune in the timeline editor.
7. Export – either a final MP4 or a CapCut draft folder.

License & usage notes

• The code is publicly available for personal learning and non‑commercial use only. Commercial deployment, resale or paid services require explicit written permission from the author.
• Users must have rights to any source media and must obtain consent before using voice‑cloning features.

---

All details above are taken directly from the repository’s README; no additional features have been inferred.