No upload · No watermark · No minute quotas
Auto subtitle generator that runs on your device
Drop a video and Whisper AI transcribes it in this browser tab — word-level timestamps, 99-language auto-detect, karaoke styling. Nothing uploads, so a 60-second reel is captioned in about 10–20 seconds instead of sitting in a render queue. Burn the captions in as MP4 or download SRT/VTT free.
- No upload — footage never leaves your device (check the Network panel)
- No minute quotas at any tier, no watermark, ever
- Word-level timing: karaoke highlight, sweep, pop and reveal styles
- Free: unlimited videos up to 3 minutes, 1080p/30 burn-in, unlimited SRT/VTT
How it works
- 1 Drop a video into the editor above — no signup, no upload.
- 2 The Whisper model loads once (size shown up front) and is cached for next time.
- 3 Captions appear with word-level timing; pick a preset and click any word to fix it.
- 4 Export a burned-in MP4, or download SRT/VTT — both free.
Frequently asked questions
How can it generate subtitles without uploading my video?
The Whisper speech-recognition model runs inside your browser on your own GPU (WebGPU). CaptionForge downloads the model once, caches it, and processes your footage locally — open DevTools' Network tab while it transcribes and you'll see zero upload traffic.
How accurate is the transcription?
It's the same Whisper model class the cloud tools use, so accuracy is comparable: very good on clean speech, weaker on noisy audio and unusual names — for every tool. The difference is the fixing workflow: low-confidence lines get a review badge, and clicking a word to correct it takes seconds, with timings redistributing automatically.
What languages are supported?
Whisper covers 99 languages with automatic detection — drop the video and the language is picked up from the audio. You can override the detection if a clip mixes languages.
What's free and what's paid?
Free: unlimited videos up to 3 minutes each, 4 caption presets including karaoke, the full basic word editor, 1080p/30 burn-in with no watermark, and unlimited SRT/VTT export. Studio ($9/mo, $69/yr or $99 lifetime) adds unlimited length, all 8 presets plus a custom style editor, brand fonts, find & replace with a custom vocabulary, SRT import, 4K/60 export and batch.
How fast is it?
On a laptop with WebGPU (Apple Silicon, recent Intel/AMD/NVIDIA), a 60-second clip transcribes in roughly 10–20 seconds and the MP4 burns in at one times realtime or faster. There is no upload wait and no server queue — those are usually the slow parts elsewhere.
Does it work offline?
After the first visit, yes — the model is cached on your device, so transcription works with the network disconnected. Try it in airplane mode; it's also the simplest way to verify nothing is being sent anywhere.