Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/KittenML/KittenTTS/llms.txt

Use this file to discover all available pages before exploring further.

KittenTTS provides three model families at different sizes, letting you trade off quality, speed, and footprint depending on your target environment.

​
Model comparison

ModelParametersSizeQualityBest for
kitten-tts-mini80M80 MBHighestProduction apps, quality-first use cases
kitten-tts-micro40M41 MBBalancedMost applications
kitten-tts-nano (fp32)15M56 MBGoodEdge devices, real-time inference
kitten-tts-nano (int8)15M25 MBGoodEmbedded systems, mobile

​
Choosing a model

kitten-tts-mini

80M params — 80 MBHighest output quality. Use when size is not a constraint and you need the best results for production applications.

kitten-tts-micro

40M params — 41 MBBalanced quality and speed. A good default for most applications where you want a smaller footprint without sacrificing too much quality.

kitten-tts-nano

15M params — 56 MB (fp32) / 25 MB (int8)Fastest inference. Best for edge devices, embedded systems, and real-time synthesis on constrained hardware.

kitten-tts-nano (int8)

15M params — 25 MBSmallest footprint via int8 quantization. Targets mobile and embedded deployments. See the nano page for known issues.

​
Switching between models

Pass the Hugging Face repo ID to KittenTTS to select a model. All models are downloaded automatically on first use. 
from kittentts import KittenTTS

# Highest quality (80M params, 80MB)
model = KittenTTS("KittenML/kitten-tts-mini-0.8")

# Balanced (40M params, 41MB)
model = KittenTTS("KittenML/kitten-tts-micro-0.8")

# Fastest (15M params, 56MB)
model = KittenTTS("KittenML/kitten-tts-nano-0.8")

# Smallest (15M params, 25MB, int8 quantized)
model = KittenTTS("KittenML/kitten-tts-nano-0.8-int8")

​
Model caching

By default, models are cached in the platform’s standard cache directory. Use the cache_dir parameter to specify a custom location, which is useful for shared environments or when managing disk space explicitly. 
model = KittenTTS("KittenML/kitten-tts-mini-0.8", cache_dir="/data/models")
Set cache_dir to a persistent volume when running in containers so the model is not re-downloaded on every restart.

Build docs developers (and LLMs) love

Get started for freeTalk to us