YouDub-webui
liuzhao1225/YouDub-webuiReviewed by AI
Checked by RepoPicks Radar on
What is YouDub-webui?
YouDub Desktop is a Windows and Mac app that turns a video into another language. You import a file or paste a link, and it writes down the speech with Whisper, translates it with a model you set up, makes subtitles, and dubs it sentence by sentence in a copy of the original voice, keeping the background sound if you want. It can give you the original audio with subtitles, the dub, or both. The steps run as a workflow you can stop, correct by hand and redo. The listening and voice work happen on your machine; the translation uses a provider key.
How easy is it to install YouDub-webui?
B · installer plus one simple step · Windows
- Windows B
- macOS B
Windows and Mac both get a B: run YouDub-windows-x64-Setup.exe or the .dmg (an Apple Silicon build; there is no Intel one), and the app prepares its own Python environment at first launch. What is left is in-app: add a provider and a key for the translation model, then let it fetch the models each workflow asks for (Whisper, Demucs, VoxCPM, the aligner). The settings file notes FFmpeg has to be a build with subtitle support, and on Windows one with the av*.dll files beside ffmpeg.exe.
YouDub-webui on GitHub
- Stars
- 5,829on Nov 10, 2026
- Downloads
- 6,961789 this week
- Growth
- +29stars in 6 days
Stars over time
Facts
- Runs on
- Windows, macOS
- You need to install
- FFmpeg
- To build from source
- Python, Node.js, NVIDIA graphics card, FFmpeg
- Disk space
- 117 MB–350 MB · from the installer
- Accounts or keys
- A translation model provider, added under Settings → Models (the example settings default to gpt-4.1-mini)
- Cost
- Free, API costs extra · The app is free under Apache-2.0 and the heavy parts (Whisper, Demucs, VoxCPM) run on your own machine. The translation step calls a model provider you configure, with gpt-4.1-mini as the default in the example settings, so a long video costs whatever that provider charges.
- License
- Apache-2.0 (Permissive)
- Last change on GitHub
- Nov 6, 2026
YouDub-webui risk signals (5)
- AttentionDependencies with known vulnerabilities
Pinned dependencies with known vulnerabilities (OSV): ajv: GHSA-2g4f-4pwh-qvx6; undici: GHSA-2m5v-pch6-74hx, GHSA-4qv8-2p9j-2685, GHSA-4w7w-5xrc-9h26; torch: GHSA-63cw-57p8-fm3p, GHSA-qfhq-4f3w-5fph, GHSA-rrmf-rvhw-rf47
automatic check
- Attentioncopies a real persons voice
The headline feature is "逐句原声克隆" — each translated line is spoken with "对应原文句子的音频片段作为参考", a clip of the original speaker as the reference. The demo on the front page does this to a filmed interview. Making someone appear to speak a language they did not speak needs their permission and is restricted in several countries.
AI review
- Attentiondownloads and re dubs other peoples videos
Link import "支持 YouTube、Shorts、B 站及 b23.tv 单视频链接;使用 yt-dlp", with no limit on length or file size, and the examples are other people's YouTube videos. Downloading from YouTube is against its terms of service, and publishing a dubbed copy of a video you do not own is a copyright matter wherever you post it.
AI review
- Infothe local interface has no password
.env.example states it plainly: "Browser session settings. No access password is required.", with YOUDUB_AUTH_COOKIE_SECURE=false. It listens on YOUDUB_HOST=127.0.0.1 by default, so only your machine reaches it — but that setting is yours to change, and nothing asks for a password if you do.
AI review
- Infoearly version with parts the author calls unverified
package.json says version 0.0.12 and the "当前范围" section lists what is not finished: "本地 workflow、Client 与 Python 插件已有验证;GitHub/npm 远端插件实装尚未验收" (remote plugins not accepted yet), no plugin market, and plugin changes only take effect after a restart. The desktop shell is built on "裁剪后的 DeepSeek Harness 框架", a fork kept inside vendor/.
AI review
Evidence is shown in its original language.
What Radar could not check about YouDub-webui
- Whether the installer brings its own FFmpeg with subtitle support or you have to install one
- How slow it is on a machine with no NVIDIA card, since the setup also accepts the processor alone
- How much disk the app takes once the models for a workflow are downloaded
- How good the translation and the sentence splitting are outside English, Chinese and Japanese
- Whether the cloned voice sounds like the demos on videos of your own