Show HN: Local AI-powered subtitle generation, translation and embedding (github.com)

A new open-source CLI called subtool (Show HN) provides local, AI-powered subtitle extraction, translation and embedding for MP4/MKV video files. It uses OpenAI’s Whisper for automatic speech recognition (auto-detects language) and Meta’s NLLB-200 for multi-language translation, producing SRT or ASS subtitle tracks (ASS recommended for proper Unicode/CJK rendering). The tool embeds multiple language tracks with correct language naming into video containers, saves subtitle files alongside the video, and supports installing via uv tooling, uvx, or pip. This is significant for AI/ML practitioners and content teams who need privacy-preserving, reproducible, offline workflows and fine-grained control over speed/accuracy trade-offs. Subtool exposes Whisper model choices (tiny → turbo → large) and NLLB translation for 12 common target languages (en, es, fr, de, it, pt, ru, ja, ko, zh, ar, hi), defaults to the turbo Whisper model, and caches downloads in ~/.cache/huggingface/ (first run downloads ~1.5GB for Whisper turbo and ~1.2GB for NLLB-200). It requires ffmpeg and supports MKV/ASS with embedded fonts for correct CJK rendering. Typical usage is simple (subtool video.mp4 -t en es), and options include explicit source language, model selection, MKV output, or SRT-only exports—making it practical for batch processing, localization pipelines, and offline research.