IndexTTS
index-tts/index-ttsReviewed by AI
Checked by RepoPicks Radar on
What is IndexTTS?
IndexTTS, from bilibili, turns written text into speech using a voice copied from a single short reference recording. You can set the emotion (from another clip, from sliders or from a description), the speaking speed and the pronunciation of specific words. It runs on your own computer through a page in the browser. You need the terminal, Python with the uv tool, an NVIDIA graphics card with CUDA 12.8 for good speed, and a few gigabytes of model files downloaded from HuggingFace or ModelScope.
IndexTTS is free and can replace ElevenLabs, Murf, and Fish Audio.
How easy is it to install IndexTTS?
D · terminal, several tools or a specific graphics card · Windows
- Windows D
- macOS D
- Linux D
On Windows and Linux, the best case, you clone the repository, install the uv tool, run `uv sync --all-extras`, install a model-download tool, fetch several gigabytes of model files and then start `uv run webui.py`; CUDA Toolkit 12.8 or newer is needed and the README warns that DeepSpeed may be hard to install on Windows. On Mac there is no CUDA support in the project's own setup, so speed depends on the processor.
Coming soon: install in 1 click with the RepoPicks app.
IndexTTS on GitHub
- Stars
- 25,032on Nov 10, 2026
- Downloads
- —all releases
- Growth
- +89stars in 6 days
Stars over time
Facts
- Runs on
- Windows, macOS, 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
- None
- Cost
- Free · No price; the license asks very large companies to request a separate licence and says to email [email protected] for commercial use
- License
- NOASSERTION (Other)
- Can replace
- ElevenLabs, Murf, Fish Audio
- Last change on GitHub
- Sep 29, 2026
IndexTTS risk signals (4)
- Attentionrestrictive custom license
LICENSE is the "bilibili Model Use License Agreement", not a standard open-source licence: a separate licence is required above 100 million monthly users or RMB 1 billion revenue, use "to improve any AI model" other than this one is forbidden, and the owner "reserve[s] the right to revoke the license"
AI review
- Attentionvoice cloning impersonation risk
The README describes zero-shot cloning of a voice from a single reference clip ("clones a voice from a single reference audio clip"), which allows copying a real person's voice without consent; the repository ships a separate DISCLAIMER file
AI review
- Infolicense forbids high risk use
LICENSE section 4.2 "Prohibited High-Risk Use" lists medical diagnosis, autonomous driving, military applications, critical-infrastructure control, large-scale biometric surveillance and automated credit or employment decisions
AI review
- Infodownloads models at run time
README: "Some small models are downloaded automatically on first run" and the example audio files are "downloaded on demand from HuggingFace/ModelScope the first time the WebUI starts"
AI review
Evidence is shown in its original language.
What Radar could not check about IndexTTS
- Real disk use and video memory needed after downloading the 2.5 and 2 model files
- Whether it installs and runs on a Mac without an NVIDIA card, and how slow it is there
- Whether the quality in each of the five languages matches the tables in the README
- Which files the automatic first-run downloads bring in