Creator tooling · Electron + Python · Personal project
Captions that transcribe, style, and animate — without a single byte leaving your machine.
A Windows desktop app that listens to a video's audio with Whisper, lets you edit and style the transcript exactly the way you want, and burns the finished captions into the file with ffmpeg — no upload, no subscription, no cloud service in the loop.
The problem
Captioning tools that are actually good all want your video uploaded to their server first.
Every serious auto-caption tool worth using is a web app or a subscription SaaS — which means the video goes up before a single word gets transcribed. For client work, personal footage, or anything you'd rather not hand to a third party, that's a hard no. The alternative was usually "learn ASS subtitle syntax and do it by hand in Aegisub," which is its own kind of painful.
What it does
Everything a paid captioning SaaS does, running entirely offline.
Local, word-level transcription
faster-whisper runs the speech-to-text entirely on your machine — GPU-accelerated when there's an NVIDIA card, falling back to CPU automatically when there isn't.
An editable transcript, not a black box
Every caption's text and timing is editable before export, and you can drag the caption itself around the preview to reposition it visually.
15 caption animations
From a simple fade to word-by-word Shorts-style pop-ins, karaoke word highlighting, typewriter reveal, and a full set of motion and glow effects — all rendered natively, not composited after the fact.
Multi-language, including translation
Auto-detect or pick the spoken language explicitly from 20+ options, and optionally translate straight to English captions using Whisper's own translation mode.
Why it holds up
Nothing ever leaves the machine
The video, the audio, the transcript — all of it stays local, start to finish. There's no server component to this app at all.
Non-Latin scripts render correctly, guaranteed
A Devanagari font ships inside the app itself, so Hindi captions render correctly even on a Windows install that doesn't happen to have a matching system font — a real failure mode I found and fixed, not a hypothetical.
Genuinely standalone
The installer bundles its own Python, faster-whisper, and ffmpeg — no separate install steps, no PATH configuration, nothing else to set up first.
Want the real build story — the ASS rendering quirks and the GPU fallback logic?
Read the full case study