# 8-cut [![License: GPL v3](https://img.shields.io/badge/License-GPLv3-blue.svg)](https://github.com/ethanfel/8-cut/blob/master/LICENSE) **Source:** https://github.com/ethanfel/8-cut A desktop tool for cutting 8-second clips from video files, designed for building [SELVA](https://github.com/google-deepmind/selva) datasets. ## Overview 8-cut lets you scrub through a video, mark a cut point, and export a fixed 8-second clip with one keypress. It tracks every export in a local SQLite database so you can resume a session or switch between resolution variants of the same source without duplicating work. All clips are exactly 8 seconds — this is a hard constraint of the SELVA format. ## Features - **Frame-accurate scrubbing** — click or drag the timeline; arrow keys and J/K/L for frame-by-frame stepping - **Keyboard shortcuts** — J/L step one frame, Shift+J/L step one second, Space/P play/pause, K pause and return to cursor, E export, M jump to next marker - **Two export formats** — H.264/AAC MP4 or lossless WebP image sequence (frames + `.wav` audio extracted alongside) - **Portrait crop** — crop to 9:16, 4:5, or 1:1 before export; adjustable horizontal crop position - **Resize** — scale short side to a fixed pixel size (e.g. 256) - **Export history** — timeline markers show previously exported clips; fuzzy filename matching detects resolution variants of the same file (e.g. `_2160p` vs `_1080p`) - **Mask generation** — generate binary foreground masks per-frame using SAM2 (segmentation) or Depth Anything V2 (depth-based), via a bundled venv - **Playlist** — drag-and-drop multiple files; duplicates are ignored ## Requirements - Python 3.11+ - `ffmpeg` in `PATH` - PyQt6 - python-mpv (requires libmpv) ``` pip install -r requirements.txt ``` For mask generation tools, additional dependencies (PyTorch, transformers, segment-anything-2, opencv) are installed into `~/.8cut/venv/` via the Settings dialog. ### Platform notes **Linux** — install libmpv via your package manager (`apt install libmpv-dev` / `pacman -S mpv`). **macOS** — install libmpv via Homebrew: `brew install mpv`. **Windows** — `python-mpv` requires `mpv-2.dll` in `PATH` or in the same directory as `main.py`. Download it from the [mpv Windows builds](https://sourceforge.net/projects/mpv-player-windows/files/libmpv/) page (pick the latest `mpv-dev-x86_64-*.7z`, extract `mpv-2.dll`). Also ensure `ffmpeg.exe` is in `PATH` (e.g. via [winget](https://winget.run/): `winget install ffmpeg`). ## Usage ``` python main.py ``` Drop a video onto the playlist or use the file picker. Scrub to your cut point, set the output folder and clip name, then press **Export** (or `E`). ### Export formats | Format | Output | |--------|--------| | MP4 | `/_NNN.mp4` — H.264 video + AAC audio | | WebP sequence | `/_NNN/frame_%04d.webp` — lossless WebP frames + `_NNN.wav` PCM audio | ### Keyboard shortcuts | Key | Action | |-----|--------| | `←` / `J` | Step back 1 frame | | `→` / `L` | Step forward 1 frame | | `Shift+←` / `Shift+J` | Step back 1 second | | `Shift+→` / `Shift+L` | Step forward 1 second | | `Space` / `P` | Toggle play/pause | | `K` | Pause and snap video to cursor | | `E` | Export clip | | `M` | Jump to next export marker (wraps) | Arrow keys and J/K/L are ignored when a text field has focus. ### Mask generation tools > **Warning:** The mask generation feature is untested and may not work reliably. For production use, consider [ComfyUI](https://github.com/comfyanonymous/ComfyUI) instead. Two standalone scripts live in `tools/`. They are run by the app via a managed venv but can also be called directly: ``` python tools/sam_masks.py --input clip.mp4 --output masks_dir/ python tools/depth_masks.py --input clip.mp4 --output masks_dir/ ``` Both output one binary PNG per frame (`frame_0000.png`, …) where white = foreground. - **SAM2** (`sam_masks.py`) — uses `facebook/sam2-hiera-large`; center-point prompt propagated across all frames - **Depth Anything V2** (`depth_masks.py`) — uses `depth-anything/Depth-Anything-V2-Large-hf`; Otsu threshold on the depth map ## Database Export history is stored in `~/.8cut.db` (SQLite). The database records filename, start time, and output path for every clip. When you open a file, 8-cut checks whether a similar filename has been processed before (stripping resolution tags like `_2160p`, `_1080p`, codec tags, etc.) and pre-populates the timeline with existing markers. ## Testing ``` pytest tests/ -v ``` 38 unit tests covering path builders, ffmpeg command generation, time formatting, and the processed-clips database. ## License [GNU General Public License v3.0](https://github.com/ethanfel/8-cut/blob/master/LICENSE)