Duix Avatar
duixcom/Duix-AvatarReviewed by AI
Checked by RepoPicks Radar on
What is Duix Avatar?
Duix.Avatar makes talking-head videos. You feed it a video of a person, it builds an avatar of their face and a copy of their voice, and from then on you type text (English, Chinese, Japanese, Korean, French, German, Arabic or Spanish) and it produces a video of that avatar speaking, with the lips matched to the sound. Everything runs on your own machine with no internet. The price is heavy hardware: Windows 10 or Ubuntu 22.04, an NVIDIA card like an RTX 4070, 32 GB of memory, Docker, and about 70 GB of downloads plus over 100 GB of free disk.
Duix Avatar is free and can replace HeyGen, Synthesia, and ElevenLabs.
How easy is it to install Duix Avatar?
D · terminal, several tools or a specific graphics card · Windows
- Windows D
- Linux D
On Windows there is a ready installer for the front end, but it does nothing on its own: first you install WSL and Docker Desktop, run docker-compose in the project's /deploy folder and wait about half an hour for roughly 70 GB of service images, with an NVIDIA card and drivers in place. Ubuntu is the same work plus installing Docker, the graphics driver and the NVIDIA Container Toolkit by hand in the terminal.
Coming soon: install in 1 click with the RepoPicks app.
Duix Avatar on GitHub
- Stars
- 16,295on Nov 10, 2026
- Downloads
- 34,387639 this week
- Growth
- +71stars in 6 days
Stars over time
Facts
- Runs on
- Windows, Linux
- You need to install
- Docker, NVIDIA graphics card
- To build from source
- Node.js, FFmpeg
- Disk space
- 193 MB–578 MB · from the installer
- Accounts or keys
- None
- Cost
- Free · No charge to download or run. The licence is the company's own "DUIX.COM Community License", not a standard open-source one, and it requires a paid commercial licence above a user limit. The company also sells a hosted avatar API as the easier route.
- License
- NOASSERTION (Other)
- Can replace
- HeyGen, Synthesia, ElevenLabs, D-ID
- Last change on GitHub
- Apr 21, 2026
Duix Avatar risk signals (7)
- Attentionnot a standard open source licence
The repository calls itself "Truly open-source AI avatar toolkit", but the LICENSE file is the "DUIX.COM COMMUNITY LICENSE AGREEMENT" and GitHub could not identify a known licence (NOASSERTION). Clause 2 says that above "1 thousand" monthly active users "you must request a commercial license from DUIX.COM, which DUIX.COM may grant to you in its sole discretion". There is also a separate "Duix.Avatar model community Licensing Agreement.pdf" covering the models.
AI review
- Attentionlicence and readme give different limits
The LICENSE sets the commercial threshold at more than 1,000 monthly active users, while the README's comparison table says "enterprises with more than 100,000 users or annual revenue exceeding 10 million USD need to sign a commercial license agreement". Anyone planning to sell videos made with it cannot tell from the repository which limit applies.
AI review
- Attentioncompany may reuse what you build
LICENSE clause 5(c): "By using the DUIX.COM Materials, you grant DUIX.COM a perpetual, worldwide, non-exclusive, royalty-free, irrevocable license to use, reproduce, modify, publicly display, publicly perform, and distribute any case studies, examples, applications, or implementations created using the DUIX.COM Materials for any purpose, including ... marketing", with no payment and no credit to you.
AI review
- Attentionmakes videos of a real person
README: it "can precisely clone your appearance and voice" from real-person video data. The same tool can clone someone else's face and voice, which is a deepfake; using a person's likeness or voice without their agreement is against the law in many countries.
AI review
- Attentionvery large download and disk use
README Windows requirements: "Must have D Drive ... More than 30GB", "C Drive ... More than 100GB", and the server install warns "download will consume about 70GB of traffic, make sure to use WiFi". Radar's 193 MB to 578 MB estimate covers only the small front-end installer.
AI review
- Infoservice images come from a separate account
README Dependencies pulls "guiji2025/fun-asr", "guiji2025/fish-speech-ziming" and "guiji2025/duix.avatar" from Docker Hub. The heavy parts that do the work are prebuilt images under a different account name than this repository, so their contents are not in the code you can read here.
AI review
- Infoquiet for months
Radar's check shows the last commit on 2026-04-21 (196 days before the check) and the last release, v1.0.6, on 2025-09-28, with 423 open issues. The README calls the community "very active" and updates "frequent", which no longer matches the repository.
AI review
Evidence is shown in its original language.
What Radar could not check about Duix Avatar
- How good the result looks: the project's own table calls the open-source lip sync a "usable effect" against the "stunning and higher definition effect" of its paid API.
- Whether it runs at all on a card other than the tested RTX 4070 and 5090, and how long one video takes.
- What is inside the three prebuilt Docker images, since they are downloaded ready-made.
- Whether it really never contacts the internet while working, as the README claims.
- The real disk use after training a model and keeping several avatars and videos.