The layers other tools make you outsource.
Schematic previews below — every one of these is live in the product today.
Erase the old captions before dubbing
OCR finds the burned-in text, an inpainting model repaints the real background underneath — clothing, water, studio walls all reconstructed. Your new captions land on clean video instead of ghost text. One pass serves every language.
One render, platform-ready looks
Pick a style at upload and every dubbed language ships with captions burned in that look — or export the SRT/ASS files and keep the styling in your NLE.
Your names stay yours in every language
One rule, every language. Lock a term once with a global default — override it per language whenever you like — and the translator obeys across every target. Each rendered line then gets machine-checked: a green ✓ names the terms it locked, an amber ⚠ flags anything that slipped. No dictionary-per-language-pair, no re-typing the same 50 terms seven times.
Two voices stay two people
Voiceprint clustering clones every participant separately and keeps identities bound through the whole video. Choose the deliverable per job:
Bring a voice bank to every job
Clone a speaker from 15 seconds of clean speech — or design a voice from a sentence ("a warm documentary narrator with a slight rasp") and freeze the one you like. Ready voices live in your library: assign per speaker, per line, or per project, and auto speaker detection fans your choice across the script.
One lane: within 12 hours.
Every job runs the identical pipeline and lands within 12 hours. Same per-line review when it arrives. Because nobody pays to keep GPUs spinning for instant turnaround, the whole realtime premium comes out of your price.