VibeRick Dub

AI dubbing agent · Built on Google Cloud

Your voice.
Every language.

Paste a YouTube link. VibeRick Dub translates each line to fit the moment it’s spoken, says it in your own voice, then listens to every line and fixes what’s wrong.

Dub a video

Same moment, same speaker. Switch languages while it plays.0:00

01

Make yours

Paste a link, choose up to four of 125 languages. The first three minutes are dubbed into each one.

The source language is skipped. A creator’s cloned voice is used only with recorded consent and a voice match; everyone else gets a matched Gemini voice. Results are public by link.

02

An agent, not a pipeline

Every line is measured, heard and repaired before it reaches you. Each step streams to the studio as it happens.

  1. Listen

    Chirp 3 timestamps every word and labels speakers. Gemini listens to the audio and corrects names and jargon; the fixes are aligned back onto Chirp’s timings.

    Chirp 3 · Gemini 3.8 Flash
  2. Brief

    Gemini writes a dubbing brief: topic, register, and how each brand and technical term is rendered.

    Gemini 3.8 Flash
  3. Translate to time

    Each sentence gets a character budget from its time slot and the voice’s measured speaking rate.

    Gemini 3.8 Flash
  4. Speak

    The creator’s own voice, if they consented and their voice matches. Other speakers are cast to the closest preset after Gemini listens to them.

    Gemini TTS voice replication
  5. Check & repair
    measureshortenGemini listensre-speakre-translateflag a human

    Too long, wrong language, missing words: the agent fixes it and listens again. What it can’t fix, it flags.

    Gemini agent loop
  6. Mix & ship

    Lines sit on the original timeline over the separated music bed, at natural speed and the voice's own level. Out come MP4, M4A tracks, SRT and VTT.

    Cloud Run Jobs · Cloud Storage

03

Measured, not promised

From real runs on a 15-minute Korean tech video and a two-speaker podcast.

366/369lines passed Gemini’s listening check on the full 15-minute video, after repairs
2:52end to end for 3 minutes into 3 languages, download to finished MP4s
125languages to dub into, each spoken by Gemini TTS in the creator’s own voice or a matched one
≈$0.03per dubbed minute in model and compute cost

05

Built on Google Cloud

One deploy, serverless end to end. Every agent action is a Firestore event the browser subscribes to.

  1. The studio posts the link to the Cloud Run API, which checks the daily quota and creates the job in Firestore.
  2. The API starts a Cloud Run Job execution for that one dub.
  3. The worker separates speech from music, then calls Chirp 3, Gemini and Gemini TTS for each language.
  4. Every agent decision is written to Firestore; the browser’s listener shows it live.
  5. Finished MP4, M4A, SRT and VTT land in Cloud Storage; a line edit re-runs only that line.

Scales by the job. Each dub is its own Cloud Run Job execution, so a hundred creators means a hundred parallel workers and an idle API costs nothing. Per-day quotas, idempotent job creation and per-job owner tokens keep the public demo safe.

06

Samples

Finished dubs. Open one to see every line, every check and every repair.

Loading…