Claude Fable 5
a3eb962b90
When yt-dlp yields no subtitle files, the server now runs scripts/transcribe.py (faster-whisper, large-v3-turbo, CUDA with CPU fallback) on the downloaded media and feeds the resulting VTT through the existing Markdown conversion. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01P1w53ybPWArpZpcei9JPYX
137 lines
4.6 KiB
Markdown
137 lines
4.6 KiB
Markdown
# Deployment
|
|
|
|
## What's in the package
|
|
|
|
```
|
|
build/ ← compiled SvelteKit server (self-contained, no npm needed)
|
|
scripts/
|
|
subtitle_to_markdown.py ← standalone subtitle converter CLI
|
|
create_zip.py ← ZIP creation helper (used internally by the server)
|
|
add_user.py ← user management CLI
|
|
start.sh ← startup script
|
|
users.txt ← user database (hashed passwords)
|
|
ads-left.html ← ad content for the left column (edit freely)
|
|
ads-right.html ← ad content for the right column (edit freely)
|
|
README.md
|
|
INSTALL.md
|
|
```
|
|
|
|
## Prerequisites on the Linux server
|
|
|
|
Node.js 18 or newer, Python 3, yt-dlp, and ffmpeg must be installed.
|
|
|
|
```bash
|
|
# ffmpeg, Node.js, and Python via apt
|
|
sudo apt install ffmpeg nodejs python3
|
|
|
|
# yt-dlp (latest release)
|
|
sudo curl -L https://github.com/yt-dlp/yt-dlp/releases/latest/download/yt-dlp \
|
|
-o /usr/local/bin/yt-dlp
|
|
sudo chmod +x /usr/local/bin/yt-dlp
|
|
```
|
|
|
|
## Installation
|
|
|
|
```bash
|
|
tar -xzf yt-dlf.tar.gz
|
|
cd yt-dlf
|
|
```
|
|
|
|
## User management
|
|
|
|
Unauthenticated visitors see the app with ad columns. Logged-in users see no ads.
|
|
The login page is at `/login` (not linked from the main page).
|
|
|
|
Add or update a user:
|
|
|
|
```bash
|
|
python3 scripts/add_user.py <username> <password>
|
|
```
|
|
|
|
Users are stored in `users.txt` (one entry per line, scrypt-hashed passwords).
|
|
Sessions are held in memory — they reset when the server restarts.
|
|
|
|
## Ad content
|
|
|
|
Edit `ads-left.html` and `ads-right.html` to replace the placeholder content with real ads.
|
|
The server caches these files at startup — restart the server after editing them.
|
|
|
|
## Running
|
|
|
|
```bash
|
|
./start.sh
|
|
```
|
|
|
|
The server listens on port 3000 by default. Open `http://your-server:3000` in a browser.
|
|
|
|
## Environment variables
|
|
|
|
Edit `start.sh` to configure the variables below, or pass them directly on the command line.
|
|
|
|
### General
|
|
|
|
| Variable | Default | Purpose |
|
|
|---|---|---|
|
|
| `PORT` | `3000` | Port to listen on |
|
|
| `ORIGIN` | `http://localhost:PORT` | Public URL of the server — set this in production |
|
|
| `YTDLP_PATH` | `yt-dlp` | Full path to yt-dlp binary if not on PATH |
|
|
| `FFMPEG_PATH` | `ffmpeg` | Full path to ffmpeg binary if not on PATH |
|
|
| `DOWNLOAD_DIR` | `~/YouTube` | Directory where downloaded files are stored on the server |
|
|
| `USERS_FILE` | `users.txt` next to `start.sh` | Path to the user database file |
|
|
| `ADS_DIR` | same directory as `start.sh` | Directory containing `ads-left.html` and `ads-right.html` |
|
|
|
|
### Whisper transcription fallback
|
|
|
|
When a download yields no subtitles (e.g. Instagram, archive.org), the server
|
|
transcribes the media locally with [faster-whisper](https://github.com/SYSTRAN/faster-whisper)
|
|
and feeds the result through the same VTT→Markdown conversion. The fallback
|
|
is skipped with a warning if faster-whisper is not installed.
|
|
|
|
Setup (GPU recommended; the pip-installed cuBLAS/cuDNN libraries are found
|
|
automatically, no `LD_LIBRARY_PATH` needed):
|
|
|
|
```bash
|
|
python3 -m venv .venv
|
|
.venv/bin/pip install faster-whisper "nvidia-cublas-cu12<13" "nvidia-cudnn-cu12<10"
|
|
```
|
|
|
|
The model (~1.6 GB for `large-v3-turbo`) is downloaded from Hugging Face on
|
|
first use and cached in `~/.cache/huggingface`.
|
|
|
|
| Variable | Default | Purpose |
|
|
|---|---|---|
|
|
| `WHISPER_PYTHON` | `python3` | Python interpreter with faster-whisper installed (e.g. `.venv/bin/python`) |
|
|
| `WHISPER_MODEL` | `large-v3-turbo` | faster-whisper model name (`large-v3` for best quality, `small` for low-end machines) |
|
|
| `WHISPER_DEVICE` | _(auto)_ | Force `cuda` or `cpu`; by default CUDA is tried first with CPU fallback |
|
|
|
|
### ZIP & Send mode
|
|
|
|
When `ZIP_AND_SEND=true`, all downloaded files are packed into a ZIP and offered as a browser download instead of (only) being saved on the server. A random prefix is added to the temporary directory name to avoid collisions when multiple users download the same video simultaneously.
|
|
|
|
| Variable | Default | Purpose |
|
|
|---|---|---|
|
|
| `ZIP_AND_SEND` | `false` | Enable ZIP & Send mode |
|
|
| `ZIP_EXTRA_DIR` | _(none)_ | Optional directory whose files are added to every ZIP alongside the downloaded files. Subdirectory structure is preserved. |
|
|
| `DELETE_AFTER_SEND` | `false` | Delete files from the server after the ZIP has been sent to the browser |
|
|
| `DELETE_DELAY` | `0` | Seconds to wait before deleting files after the ZIP has been sent |
|
|
|
|
### Examples
|
|
|
|
Custom port and public URL:
|
|
|
|
```bash
|
|
PORT=8080 ORIGIN=http://myserver:8080 ./start.sh
|
|
```
|
|
|
|
ZIP & Send with cleanup after 60 seconds:
|
|
|
|
```bash
|
|
ZIP_AND_SEND=true DELETE_AFTER_SEND=true DELETE_DELAY=60 ./start.sh
|
|
```
|
|
|
|
ZIP & Send with extra files bundled into every download:
|
|
|
|
```bash
|
|
ZIP_AND_SEND=true ZIP_EXTRA_DIR=/opt/yt-dlf/extras ./start.sh
|
|
```
|