OpenSubs

Transcription in the browser

Whisper runs as ONNX weights through transformers.js, on WebGPU where available and on WebAssembly otherwise. Models between 40 MB and 250 MB download once and stay cached afterwards.

Translation into 20 languages

Chrome's on-device translation handles captions at no cost, a bring-your-own Claude, OpenAI or DeepL key covers longer dialogue, and a paid backend option needs no third-party key.

Twelve caption presets

Core and advanced packs cover interview-clean, fitness bold, broadcast yellow, neon gaming, CJK, podcast bars and vertical safe-area layouts, each previewed over the user's own footage.

Burned-in MP4 export

libass, compiled to WebAssembly, draws the captions while WebCodecs encodes the frames, producing an MP4 whose subtitles are part of the picture and survive later re-encoding.

SRT, VTT and ASS export

Timed caption files export in three formats for other editors and players, using the same cue segmentation and line breaking that the burned-in render relies on.

Word-by-word highlighting

Captions can emphasise each word as it is spoken, with growth and glow effects, driven by the timing data produced during transcription.

One engine across four surfaces

The same AGPL-3.0 Rust engine ships as a web app, a Chrome, Edge and Firefox extension, a desktop app for macOS, Windows and Linux, and a command-line tool.

• Short-form social video: creators caption clips with word highlighting, then pick Shorts Safe Area or Reel Box so text clears the player interface.
• Interviews and documentaries: a talking-head recording is transcribed locally, cleaned line by line, then exported as a broadcast preset or editable SRT.
• Multilingual reach: one recording is translated into 20 languages, and bundled fonts make Chinese, Japanese and Arabic captions burn correctly.
• Confidential footage: material that cannot be uploaded gets subtitled entirely on the device, with no account, queue or file-size cap.
• Batch and automated work: the command-line tool and desktop app run the same engine over many files in scripts or CI.

Is the video uploaded to a server?

No. OpenSubs reads the file directly in the browser through an object URL, and transcription, styling and burning all run on the local machine. The product has no upload endpoint at all, so footage is never sent anywhere. Only subtitle text or audio can leave the device, and only when a cloud translation option is deliberately selected.

Does the tool add a watermark?

No. A burned MP4 carries nothing but the captions that were chosen, and exported SRT, VTT and ASS files are plain subtitle files. The reason is structural: the work runs in the user's browser rather than on a server that someone has to pay for.

Do I need an account or payment?

No account and no sign-in are required for transcription, on-device translation, styling, trimming, burning or any export format. The only paid option is translation through the OpenSubs backend, bought with credits at $5 per 1,000 credits, where the price of a job is shown before it runs and a failed job costs nothing.

How are subtitles generated from speech?

Whisper runs locally as ONNX weights driven by transformers.js, using WebGPU where available and WebAssembly otherwise. Models range from 40 MB to 250 MB, are downloaded once and then cached by the browser. Spoken language is detected from the audio every four seconds, so a video that switches between two languages is transcribed in both.

Which languages can captions be translated into?

Twenty, including Simplified and Traditional Chinese, Japanese, Korean, Spanish, Portuguese, French, German, Italian, Russian, Arabic, Hindi, Indonesian, Thai, Vietnamese, Turkish, Polish, Dutch and Ukrainian. Fonts for non-Latin scripts are bundled, so Chinese and Arabic subtitles render as text rather than empty boxes.

Does OpenSubs work offline?

Largely, after a first visit. The speech model is cached by the browser once downloaded, and transcription, styling, burning and every export run locally without a connection. Cloud translation is the one feature that needs the network, unless a personal API key is used instead.

Is the project open source?

Yes, under AGPL-3.0. Engine, web app, browser extension, desktop app and command-line tool live in one repository, and the same Rust engine compiles to WebAssembly for the browser and to native binaries elsewhere, which is why output is identical across all of them.

• Open the page and drop in a video. The clip is read straight off the disk inside the tab, so nothing is uploaded and no account is needed.
• Alternatively open an existing .srt or .vtt file. Existing subtitles skip transcription and go directly to styling, translation and export.
• Choose where transcription runs. On-device Whisper is free and keeps audio local; a personal Claude, OpenAI or DeepL key suits long recordings or slower machines.
• Pick a model and the spoken language. Tiny, Base and Small cover 40 MB to 250 MB, while leaving detection on samples the audio every four seconds.
• Edit any line that came out wrong. Caption text is correctable in place before styling, so transcription errors can be fixed without re-running the whole job.
• Select one of the 12 caption presets and tweak font, colour, outline, shadow or position. Each preset previews over the actual footage rather than a mock-up.
• Export the result. Burn the captions into an MP4, keeping the original audio untouched, or download SRT, VTT and ASS files for use elsewhere.
• For long files, batches or videos already playing in a tab, install the desktop app, browser extension or command-line tool built on the same engine.