RepoPicks Radar

Updated every day around 23:30 UTC

Guides

OpenTalking

datascale-ai/opentalkingReviewed by AI

Checked by RepoPicks Radar on

What is OpenTalking?

OpenTalking puts together the parts of a talking digital person: a web page where you pick an avatar and a voice, speech recognition, a chat model, text to speech and lip-sync video sent to the browser. It also makes videos from text or audio and can copy a face from a camera or an uploaded video. It is aimed at people comfortable with a terminal: you clone the repository, install Python, Node.js and FFmpeg, fill in a settings file and start the services. A test mode runs without a graphics card; real lip-sync needs an NVIDIA GPU and downloaded model files.

OpenTalking is free and open source (Apache-2.0) and can replace HeyGen, Synthesia, and D-ID.

How easy is it to install OpenTalking?

D · terminal, several tools or a specific graphics card · Windows

  • Windows D
  • macOS D
  • Linux D

Linux is the smoothest path and is still a D: there is no installer, you run shell scripts or Docker Compose, install Python, Node.js and FFmpeg, edit a settings file, and need an NVIDIA GPU plus downloaded weights for real lip-sync. Windows is described through WSL2, and on Mac the docs only cover an Apple Silicon CPU/MPS path for one model.

Coming soon: install in 1 click with the RepoPicks app.

OpenTalking on GitHub

Stars
3,344on Nov 10, 2026
Downloads
1645 this week
Growth
+35stars in 6 days

Stars over time

3,3103,3203,3303,3403,350 Nov 5Nov 6Nov 7Nov 8Nov 9Nov 10 3,3103,3203,3303,3403,350 Nov 5Nov 8

Facts

Runs on
Windows, macOS, Linux
You need to install
Python, Node.js, Docker, NVIDIA graphics card, FFmpeg
To build from source
Python, Docker, NVIDIA graphics card, FFmpeg
Disk space
1.4 GB–5.5 GB · estimate
Accounts or keys
Chat model (OpenAI-compatible, default DashScope) — OPENTALKING_LLM_API_KEY, DashScope speech recognition — OPENTALKING_STT_DASHSCOPE_API_KEY, Xiaomi MiMo speech recognition — OPENTALKING_STT_XIAOMI_API_KEY
Cost
Free · The code is free (Apache-2.0). The example settings point the chat model and speech recognition at paid cloud services (DashScope, Atlas Cloud), and the project also documents local speech and lip-sync options, so the running cost depends on which ones you choose.
License
Apache-2.0 (Permissive)
Can replace
HeyGen, Synthesia, D-ID
Last change on GitHub
Sep 4, 2026

OpenTalking risk signals (5)

  • Attentionface and voice cloning

    README has a "C. Video Clone" section with "Realtime camera imitation" and "Uploaded video imitation", plus "Cloned voice driven" video creation and Xiaomi MiMo "voice clone profiles" in .env.example. The same features can be used to make a real person appear to say things they never said; the roadmap still lists "licensed voices, and synthetic-content labeling" as not done.

    AI review

  • Attentionpackages from third party mirror

    pyproject.toml sets a non-default package source as the default: "[[tool.uv.index]] url = \"https://pypi.tuna.tsinghua.edu.cn/simple\" default = true", so installs pull Python packages from a mirror in China instead of the official PyPI unless you change it.

    AI review

  • Infoservice open to local network

    .env.example ships OPENTALKING_API_HOST=0.0.0.0 and OPENTALKING_WEB_HOST=0.0.0.0, and docker-compose.yml publishes ports 8000 and 5173, so the API and web page answer anyone who can reach your machine; no login for the web page is described.

    AI review

  • Infoimports personal chat history

    README "Recent Progress" entry for 2026-06-25: "Added WeChat memory persona import", so the avatar's memory can be fed with your private WeChat conversations.

    AI review

  • Infovery new project

    The repository was created on 2026-04-16 and has a single release, v0.1.0 (2026-06-15), with no platform binaries; pyproject.toml also says version 0.1.0.

    AI review

Evidence is shown in its original language.

What Radar could not check about OpenTalking

  • Whether real-time conversation actually keeps up on the listed graphics cards, or on Windows via WSL2 and on Apple Silicon.
  • How much disk the model weights really take after download (the checks only estimate 1.4-5.7 GB).
  • How much the cloud chat and speech services cost per hour of conversation.
  • What the third-party Compshare image linked in the README contains.
  • Whether the lip-sync and voice quality in the demo videos matches what you get on your own machine.

Similar projects

The RepoPicks app

Coming soon: install in 1 click with the RepoPicks app.

Before you download

See the release page on GitHub