Desktop subtitle factory: Whisper + your choice of translator

What it does

SmartSub (“妙幕”) is a desktop app that batch-generates subtitles from video or audio using OpenAI’s Whisper models, then translates those subtitles through a wide menu of services—cloud APIs, local Ollama instances, or OpenAI-compatible endpoints like DeepSeek. It keeps everything on your machine, which means no upload queues and no privacy negotiations with a SaaS provider.

The interesting bit

The project is essentially a polished GUI and distribution layer around whisper.cpp, but the packaging matters: it auto-detects your GPU and offers one-click CUDA or Core ML acceleration downloads, and it recently added a UI for tweaking AI model parameters (headers, request bodies, JSON objects) without touching code. That’s the kind of “boring” config work that saves hours when you’re iterating on translation prompts.

Key highlights

• Runs on Windows, macOS, and Linux; Apple Silicon gets Core ML automatically, NVIDIA gets optional CUDA boosters
• Translation backends include Baidu, Volcano, Microsoft, DeepLX, Ollama, DeerAPI, and generic OpenAI-style APIs
• Supports custom subtitle output formats (translation-only or bilingual)
• Can import existing subtitle files for translation without re-transcribing
• Model management UI with manual import for users struggling with large downloads

Caveats

• DeepLX “easily gets rate-limited during batch translation,” per the README itself
• GPU acceleration can cause crashes; the suggested fix is trial-and-error with different CUDA versions or turning it off
• Intel Macs get no hardware acceleration at all

Verdict

Worth a look if you regularly subtitle long videos and want one tool that handles transcription, translation, and format tweaking without leaving your desktop. Skip it if you need a simple one-off job—there’s still model downloading, API key wrangling, and occasional GPU debugging to get through.