Linly-Talker
Kedreamix/Linly-TalkerReviewed by AI
Checked by RepoPicks Radar on
What is Linly-Talker?
Linly-Talker is a web page you run on your own computer that puts several open models together: you upload a photo, speak or type, an AI model answers, text-to-speech reads the answer and a lip-sync model (SadTalker, Wav2Lip, MuseTalk or ER-NeRF) makes the picture talk. It can also copy a voice from about a minute of recorded speech. It is for people comfortable with the terminal: Python, an NVIDIA card, ffmpeg and several gigabytes of model files. Free, MIT licensed, and the author has moved on to a newer repository.
Linly-Talker is free and open source (MIT) and can replace HeyGen, Synthesia, and D-ID.
How easy is it to install Linly-Talker?
D · terminal, several tools or a specific graphics card · Windows
- Windows D
- Linux D
Both platforms are terminal work with an NVIDIA card: a conda environment, PyTorch matched to your CUDA version, ffmpeg, several `mim install` steps, pytorch3d from source for the NeRF part, then gigabytes of model files to download and move into the right folders. The README itself warns "The installation process is very slow". Windows is slightly helped by an "all-in-one package", but it is still a sequence of conda commands, so both stay at D.
Coming soon: install in 1 click with the RepoPicks app.
Linly-Talker on GitHub
- Stars
- 3,481on Nov 10, 2026
- Downloads
- —all releases
- Growth
- +2stars in 6 days
Stars over time
Facts
- Runs on
- Windows, Linux
- You need to install
- Python, NVIDIA graphics card, FFmpeg
- To build from source
- Python, NVIDIA graphics card, FFmpeg
- Disk space
- 1,010 MB–4.5 GB · estimate
- Accounts or keys
- Google Gemini (only if you pick GeminiPro as the chat model), OpenAI (only if you pick ChatGPT as the chat model)
- Cost
- Free · MIT licence, no payment. The chat part can use a free local model (Qwen, Linly) or an online one such as Gemini or ChatGPT, which you pay for.
- License
- MIT (Permissive)
- Can replace
- HeyGen, Synthesia, D-ID, ElevenLabs
- Last change on GitHub
- Feb 10, 2026
Linly-Talker risk signals (7)
- Attentionstale releases
Last release on 2024-04-23 (925 days ago)
automatic check
- AttentionDependencies with known vulnerabilities
Pinned dependencies with known vulnerabilities (OSV): joblib: GHSA-6hrg-qmvc-2xh8, PYSEC-2022-288; gradio: GHSA-26jh-r8g2-6fpr, GHSA-279j-x4gx-hfrh, GHSA-37qc-qgx6-9xjv; transformers: GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm, GHSA-37q5-v5qm-c9v8; protobuf: GHSA-7gcm-g887-7qv7, GHSA-8qvm-5x2c-j2w7, PYSEC-2026-1805; gradio: GHSA-26jh-r8g2-6fpr, GHSA-279j-x4gx-hfrh, GHSA-34rf-p3r3-58x2; scikit_learn: GHSA-jw8x-6495-233v, PYSEC-2024-110; protobuf: GHSA-7gcm-g887-7qv7, GHSA-8qvm-5x2c-j2w7, PYSEC-2026-1805; basicsr: GHSA-86w8-vhw6-q9qq, PYSEC-2026-1215; diffusers: GHSA-7wx4-6vff-v64p, GHSA-98h9-4798-4q5v, PYSEC-2026-2446; accelerate: GHSA-4j2p-28q2-5m79, PYSEC-2026-3804; transformers: GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm, GHSA-4w7r-h757-3r74; modelscope: GHSA-fhhq-h4hg-549x, PYSEC-2026-2663; lightning: GHSA-cgwc-qvrx-rf7f, GHSA-mr7h-w2qc-ffc2, GHSA-qqmf-gpg7-g8gw
automatic check
- Attentionmakes a real person say things
The project clones a voice from "a one-minute voice sample" and animates any uploaded photo; the README's own section title is "Voice Clone - Stealing Your Voice Quietly During Conversations", and the ER-NeRF part says "Taking inspiration from the likeness of Obama". Using someone else's face or voice this way can break the law and the rules of most platforms.
AI review
- Attentionunofficial access to paid ai services
Both requirement files install "g4f" (GPT4Free) and the README has a "GPT4Free" entry among the chat models. That library reaches paid providers' chat services without an account, which is against those providers' terms.
AI review
- Attentionmodels from file sharing links
Model files and the "Windows All-in-One Package" are offered through Baidu Pan (password `linl`), Quark drive (pan.quark.cn) and an Aliyun Docker image, alongside Hugging Face and ModelScope. Files from those drives come with no checksum, so you cannot tell what you are running.
AI review
- Infodevelopment moved on
The 2026.02 update points to a different repository, Linly-Talker-Stream, as "A Brand New Real-time Streaming Interactive Architecture"; the last commit here is 2026-02-10 and the only release is v1.0 from April 2024.
AI review
- Infoone voice feature already broken
README warning about Edge TTS: "it seems that Microsoft has restricted certain IPs... The solution is still unstable for now", with a recommendation to use CosyVoice instead.
AI review
Evidence is shown in its original language.
What Radar could not check about Linly-Talker
- Whether the long install still works, given pinned packages from 2024 and dependencies pulled from many sources
- How much video memory each avatar model really needs and how long one video takes
- Real disk use after downloading the chat, voice and avatar models
- Whether the Colab notebook in the repository still runs
- What the Windows all-in-one package on the Quark drive contains