Paste a YouTube link or open a file. Translay transcribes it, translates it and dubs it with each person's cloned voice, in sync with the picture. All on your computer, with your GPU.
Across 12 test videos with 2 to 4 people, with and without noise. Speed, on a podcast with café noise.
While you watch, the aligned transcript shows who is speaking, what they said and its translation. Switch back to the original voice any time.
It happens while you watch, starting where you are. Skip ahead and the work skips with you.
Exactly, that's why every seed counts. Let's call it a day.
Exacto, por eso cada semilla cuenta. Por hoy terminamos.
A complete video library, built around what you add, what you watch and what you want to take with you.
YouTube videos, shorts and streams, playlists that update themselves, a CSV with thousands of links, Dropbox or files dropped on the window. You can also record your screen and translate it.
It lives on the server, even if you close the tab. First what you are watching, then the rest.
Titles, channels and every sentence, in the original and the translation, ignoring accents.
Without gravity, how do the roots know which way to grow?
What the video was about and all its key points, made on your computer. One click jumps to that minute.
Name each one and choose their voice: their own, another person's from the video or one saved in the library.
Terms that are never translated or always translated the same way. Change it and what's done is re-translated.
From the player, a card, a whole collection or a selection. If the video isn't ready yet, the export waits.
Automatic collections per channel and category, with the channel's real avatar and banner. Each library in its own folder, another drive included.
For classes and meetings. Record or upload the video and use its transcript and summary, in its language, with no translation or voice.
What you delete comes back exactly as it was, with its translations, voices and collections.
Real screenshots of the app, with public-domain videos from NASA and NOAA.
The dubbed video with subtitles in both languages and the aligned transcript.
The original and the translated side by side, sentence by sentence.
Continue watching and a row per collection, each with its banner.
Who speaks in the video and the voice each one uses.
Downloads and processing with their state and progress.
Several links at once, straight into a collection or category.
Video, audio, subtitles and transcript in one step.
Videos, collections, channels and everything said, in separate groups.
Several libraries, each with its own folder and banner.
Every option with its description, applied instantly.
On startup, Translay looks at your GPU, memory and processor and picks the most complete setup you can watch live.
A 10 min video takes ~5 min
Models use the GPU only while processing, so you can game or edit afterwards.
VRAM, RAM and CPU threads, to leave room for other programs. RAM and threads apply without restarting.
If the profile is too much for your machine, memory runs short or another program is using the GPU, it explains it and suggests what to do.
Nothing leaves your computer except what you download from YouTube. And you can open it from other devices on your network.
On your network you sign in with the owner's password. Each library can be private, password-protected or public, with its own link.
Transcripts, voices and downloads live in each library's folder, written with a backup copy.
Everything the app does, also over HTTP, with per-library keys and signed notifications when something finishes.
GET /api/v1/videos?q=raíces
Authorization: Bearer ••••POST https://…/webhook
{ "event": "video.procesado" }The Windows installer sets everything up by itself, no commands needed.
git clone https://github.com/alenj0x1/translay.git && cd translay
cp .env.example .env # and change TRANSLAY_OWNER_PASSWORD
docker compose up -d --buildOpen http://localhost:8765. The first time it downloads the models (~6 GB).
git clone https://github.com/alenj0x1/translay.git && cd translay
pip install -r requirements.txt
uv venv -p 3.11 .venvs/omnivoice
uv pip install -p .venvs/omnivoice/Scripts/python.exe --torch-backend=cu128 omnivoice deepfilternet soundfile
python run.pyYou need Python 3.11+, uv, ffmpeg and Node.js 20+. Open http://127.0.0.1:8765.
Released on November 13, 2026.
Each person in a video has their own muted color, on their name in the transcript and on their P1, P2…, so you can follow who speaks at a glance
The add box comes in with a soft animation when you arrive, and its title changes with a short animation when you pick «Traducir y doblar» or «Solo transcribir»
A new home page that presents Translay, with a playable sample of the player and its aligned transcript, the five steps of dubbing, every feature, real screenshots, the profile picked for each GPU, access and API, installation, what's new in the latest version and frequent questions, in English and Spanish and in light and dark
No. Translay is free and open source, under the MIT license. The models it uses keep their own licenses.
Only to download YouTube videos and the models the first time (about 6 GB). Translating, dubbing and summarizing happen on your computer.
Yes, with subtitles, transcripts and summaries. The cloned voice needs an NVIDIA GPU, from 4 GB with lighter models.
From English and Portuguese into Latin American Spanish. Any video can also be only transcribed, in its own language.
About 7.5 GB for the engine and 6 GB of models, plus your videos. The Storage screen shows what can be freed.
Yes, from source or with Docker and an NVIDIA GPU. Then open it from the browser of any device on your network.
Yes. An API with per-library keys lets you add, search and export, and webhooks tell you when a video, an export or the queue finishes.
Yes. The translated video comes out as an MP4 with the Spanish voice and the original as a second track, plus subtitles, audio and text.