KittenTTS 2 Brings Local Voice Cloning

KittenTTS 2 makes local voice cloning substantially more practical, moving a capability usually associated with hosted APIs onto developer-controlled hardware. The important caveat is licensing: the code is Apache-2.0, but the weights carry commercial thresholds and attribution obligations.

• –CPU support and a llama.cpp fork reduce deployment friction, although 1.7B parameters make it much heavier than KittenTTS’s original 15M–80M models. [GitHub README](https://github.com/KittenML/KittenTTS)
• –In-context cloning eliminates fine-tuning, enabling private narrators, accessibility tools, game characters, and local voice agents.
• –Built-in voices, multilingual support, and inline emotion controls make it more useful than basic waveform-generation libraries.
• –Organizations exceeding $1 million in annual revenue or cumulative funding need a separate license, and distributions require “Powered by Stellon Labs” attribution. [License](https://huggingface.co/KittenML/kitten-tts-2/blob/main/LICENSE.md)