# Video Toolkit - Python Dependencies
# Install: pip install -r requirements.txt
#
# Core (required for all features)
requests>=2.28.0
yt-dlp>=2024.1.0

# Speech-to-Text (for extract-text / video transcription)
# Vosk is the DEFAULT offline engine: pure C++ inference, no torch /
# no ctranslate2, so it runs everywhere (including sandboxed environments
# where openai-whisper / faster-whisper fail to load). The Chinese model
# (vosk-model-small-cn-0.22) is auto-downloaded on first use.
vosk>=0.3.45

# Optional HIGHER-ACCURACY engine (auto-used when installed).
# Needs PyTorch and is slower on CPU — only install if you want the
# best possible transcript accuracy:
# openai-whisper>=3.0.0        # engine=auto/whisper
# faster-whisper>=1.0.0       # ctranslate2 — NOTE: segfaults on Python 3.13

# OCR (for image-text extraction)
# PaddleOCR is recommended: best Chinese text recognition
paddleocr>=2.6.0
paddlepaddle>=2.4.0

# Image processing (for subtitle removal via inpaint)
opencv-python>=4.7.0
numpy>=1.24.0
Pillow>=9.4.0

# Fallback options (only install if primary fails):
# easyocr>=1.6.0             # Alternative to PaddleOCR
# pytesseract>=0.3.10        # Requires tesseract-ocr system package
