Under the hood: ffmpeg for media plumbing, OpenAI Whisper for transcription, GPT-5.4 for personality-preserving translation, and libx264 for the final burn-in.
Drag a video/audio file (up to 5 GB) or paste text directly. LinguaSub streams the upload to disk so browser tabs stay responsive even for hour-long files.
Any of the 21 supported languages either side. A one-click swap button flips them — 'Hindi → English' becomes 'English → Hindi' instantly. A guard prevents same-source-same-target.
For audio/video, ffmpeg extracts a compressed 16 kHz mono track, splits it into ≤ 24 MB chunks, and OpenAI Whisper transcribes each chunk with segment timestamps offset back to the original timeline.
The transcript is sent to GPT-5.4 with a subtitle-specific prompt: preserve tone, register, humour, sarcasm, honorifics, Bollywood expressions. Long transcripts are batched (60 segments per call) so an hour-long clip still finishes in a couple of minutes.
The Edit & Render panel shows the video with each segment as a clickable card. Click a line → the video pauses and jumps to that segment. Edit the translated text, tweak start/end. Save whenever you want.
One click triggers a background ffmpeg job that burns subtitles into the frames (libx264, white text with black outline, 40 px margin). The frontend polls a status endpoint so you know when it's ready.
Three files ready: the original SRT for YouTube Studio, the VTT for web players like Vimeo, and the burned-in MP4 for Reels, TikTok, or WhatsApp forwards.
Every job runs through a purpose-built pipeline. Each layer is visible in the app so you can see exactly what changed.
Detect speech and speakers — figure out who is talking so translations can hold voice across turns.
Generate an accurate transcript in the source language with word-level timestamps.
Extract characters, relationships, scene and emotion so the translation reads in-character.
Turn source lines into natural, native-sounding target lines using the story brief and speaker tags.
Adapt slang, humour, idioms and cultural references so jokes actually land.
Optimise line breaks and reading speed — cap at 42 chars per line, 17 chars per second.
Verify timing, grammar, terminology consistency and enforce the caption spec end-to-end.
A 5-minute video: ~30 seconds transcription + ~15 seconds translation + ~20 seconds burn-in ≈ 1 minute total. A 60-minute podcast: ~5 minutes transcription + ~1 minute translation + ~4 minutes burn-in ≈ 10 minutes.
The upload and processing happen server-side, so once the upload completes you can close the tab. Come back later — your job is in the History panel with the SRT/VTT ready.
Yes. You can save edits and download the updated SRT/VTT without triggering a burn-in. Render is only needed if you want the finished MP4.
That's what the editor is for. Every segment is editable in place — the translated text, the start time, and the end time. Save your fix, and the SRT/VTT regenerate instantly.
Web UI is one file at a time right now. A public API is on the roadmap for Studio and Agency plans.
One-time free trial for new users. Upload a clip and watch each step happen live.