KittenTTS
KittenML/KittenTTSReviewed by AI
Checked by RepoPicks Radar on
What is KittenTTS?
KittenTTS reads text aloud with models that run on your own machine, no account and no internet call needed once the files are downloaded. The main model has 1.7 billion parameters, 115 voices and 20 languages, and can copy a speaker from 5 to 30 seconds of audio; there are also tiny models of 15 to 80 million parameters for weak computers. It is a Python library: you install it with pip and write a few lines of code, or start the included Gradio page for a simple screen. A graphics card is optional, but about 6 GB of free memory is not.
KittenTTS is free and open source (Apache-2.0) and can replace ElevenLabs, Murf, and Speechify.
How easy is it to install KittenTTS?
C · install 1 or 2 tools · Windows
- Windows C
- macOS C
- Linux C
Grade C on all three systems: with Python 3.10 or later you run "pip install kittenml", then either write a few lines of Python or run "pip install kittenml gradio" and "python gradio_demo.py" for a local page. Windows may also need Microsoft's Visual C++ runtime, the faster Mac path means installing the extra MLX backend, and the NVIDIA route on Linux asks for driver 580 or newer, a C compiler and Python development headers.
KittenTTS on GitHub
- Stars
- 16,831on Nov 10, 2026
- Downloads
- 3,511,934186,516 this week
- Growth
- +37stars in 6 days
Stars over time
Facts
- Runs on
- Windows, macOS, Linux
- You need to install
- Python
- To build from source
- Python, NVIDIA graphics card
- Disk space
- 1.2 GB–5.1 GB · estimate
- Accounts or keys
- None
- Cost
- Free · The code is Apache-2.0 and the models are downloaded at no charge. The same company sells "integration assistance, custom voice development, and enterprise licensing" and runs a hosted API at platform.kittenml.com, described in the README as free.
- License
- Apache-2.0 (Permissive)
- Can replace
- ElevenLabs, Murf, Speechify
- Last change on GitHub
- Oct 27, 2026
KittenTTS risk signals (7)
- Attentionthe models are not under the code license
README, License: "That covers the code in this repository. The models are licensed separately and their terms may differ. Each model repository carries its own licensing ... do not assume the code's license extends to the weights." KittenTTS 2 is "released under the Stellon Labs Community License", a licence written by the company, so the Apache-2.0 permission the checks found applies to the code only.
AI review
- Attentioncloning a voice needs that persons consent
README, Features: "Voice cloning -- Clone any speaker from 5-30 seconds of audio, no fine-tuning", and the measured similarity for "cloning an unseen recording" is reported as 0.81. Copying someone's voice without their permission runs into voice and image rights in many countries, and the repository sets no rule about whose recording you may use.
AI review
- Infolicense fields inside the project disagree
setup.py declares "License :: OSI Approved :: MIT License" while LICENSE is the Apache License 2.0 and pyproject.toml says license = {text = "Apache 2.0"}. The two packaging files describe different licences for the same code.
AI review
- Infowatermarking library among the dependencies
The dependency lists in pyproject.toml, requirements.txt and setup.py include "resemble-perth", the Perth audio watermarking library, grouped under the comment "s3gen vocoder (vendored in kittenml/kittentts2/s3gen.py)". The README never mentions a watermark, so whether the audio you generate carries one is not explained.
AI review
- Infoparts are labelled alpha or beta
setup.py carries "Development Status :: 3 - Alpha"; the expression controls are marked "Beta. Emotion control steers delivery rather than guaranteeing it"; and the fast C++ path says "Kitten-TTS 2 C++ is in beta". The README also notes the model "truncates or drifts into repetition when asked for a whole script in one pass".
AI review
- Infomodels are downloaded at first use
The weights live in separate HuggingFace repositories ("KittenML/kitten-tts-2", 492 MB, with a lossless 954 MB option) and the prerequisites ask for "About 2 GB of disk space for the model and audio decoder, plus Python dependencies". The program fetches those files from the internet the first time you run it, and the checks estimate 1250 to 5250 MB in total.
AI review
- Infogithub releases lag behind the code
The newest GitHub release is 0.8.1 from 2026-02-24, while pyproject.toml and setup.py are at version 0.9.7 and the last commit is 2026-10-27. What "pip install kittenml" gives you is not described by any release note here.
AI review
Evidence is shown in its original language.
What Radar could not check about KittenTTS
- Whether the voices and the cloning really sound as reported, since the numbers come from the project's own measurements
- What the Stellon Labs Community License allows, as its text lives in the model repository and not here
- Whether generated audio carries a watermark, which the dependency on resemble-perth raises but the README does not address
- Real disk and memory use on your machine, beyond the "about 2 GB" and "about 6 GB of RAM" the README states
- How fast it is on your processor, as the claim of running in real time on a CPU is the project's own
- What the hosted API at platform.kittenml.com records about the text you send it