An open-weight, 82-million-parameter text-to-speech model that runs high-quality narration locally on a CPU — free, offline, and Apache-2.0 licensed for commercial use.
Last verified · 2026-07-07 · by Moe Ameen
Kokoro TTS is an open-weight text-to-speech model released under the name Kokoro-82M by a developer who goes by hexgrad. Its defining trait is size: at roughly 82 million parameters it is a fraction of the scale of most competitive voice models, yet it produces narration clean enough that it reached the top of the TTS Spaces Arena leaderboard when its early v0.19 build appeared in December 2024. The v1.0 release followed on January 27, 2025. The weights ship under the Apache 2.0 license, which means you can run it, modify it, and use its output commercially for free.
Architecturally it builds on StyleTTS 2 with an ISTFTNet decoder and no diffusion step, which is a large part of why it is so fast. On a GPU it generates speech many times faster than real time; on an ordinary CPU it still runs at or near real time, which is unusual for a model this natural-sounding. Because there is no cloud dependency, it runs fully offline — and thanks to community ports like kokoro-js and Transformers.js, it will even run 100% locally inside a web browser via WebGPU or WebAssembly. Notably, the model was reportedly trained on only a few hundred hours of permissively licensed and synthetic audio for around a thousand dollars of GPU time, which made its quality-to-cost ratio a talking point in the open-source community.
Kokoro v1.0 ships 54 preset voices spanning several languages — American and British English, Spanish, French, Hindi, Italian, Portuguese, Japanese, and Chinese — organized by language and gender, and you can blend voices to create new ones. What it does not do natively is clone a specific person's voice from a sample; it works from its bundled voice packs (third-party projects add cloning on top). The honest framing: Kokoro is a voice model, not a product. It has no polished app, no scheduler, and no content pipeline — it turns text into an audio file, and everything around that is yours to build or bring.
Kokoro's whole appeal is that the audio layer becomes free and local: no per-character API meter, no cloud round-trip, a voice track you can generate on a laptop CPU as many times as you like. That makes it a natural fit for a faceless, narration-first channel — but a voice model on its own gives you an MP3, not a channel. It writes no script, cuts no video, makes no thumbnail, and posts nothing. Kompozy is the engine that supplies what Kokoro reads and ships everything Kokoro can't touch. Kompozy generates the Blog Article, the Text Posts, the video script, and the Email Newsletter under a Persona Brief and banned-word filters, so the words you feed into Kokoro are already on-brand and audience-fit — then it fans that same source into the visual and video formats and publishes the whole set.
The concrete pipeline: you draft or record one idea, Kompozy writes the long-form article and a tight narration script held to your voice, and you run that script through Kokoro locally to get a clean voiceover at zero marginal cost. Kompozy takes the same source and produces the pieces the model can't — Clipped and Persona Shorts for the feeds, brand-exact Carousel Posts, Quote Graphics, Photo Posts, plus the blog and newsletter — and schedules and publishes the lot across nine social platforms plus blog and email from one queue with Autopilot. For talking-head social video Kompozy already generates avatar clips with HeyGen's native TTS, so the two rarely overlap: Kokoro owns your standalone, free audio track; Kompozy owns generation and cross-platform distribution.
Kokoro TTS is an open-weight, roughly 82-million-parameter text-to-speech model (Kokoro-82M) released by the developer hexgrad under the Apache 2.0 license. It turns text into natural-sounding narration, runs fast even on a CPU, and can run fully offline — including inside a browser via community ports. Its v1.0 released on January 27, 2025.
Yes. The model weights are released under the Apache 2.0 license, so you can run Kokoro locally and use its audio output for free, including commercially. You only pay if you choose a hosted API provider that runs it for you (several offer it at low per-character rates); running it yourself on your own hardware costs nothing beyond compute.
Yes. Kokoro is built on a lightweight StyleTTS 2 / ISTFTNet architecture and runs at or near real time on an ordinary CPU, with no GPU required. Community JavaScript ports (kokoro-js, Transformers.js) also run it 100% locally in the browser using WebGPU or WebAssembly.
Not natively. Kokoro v1.0 ships 54 preset voices across several languages that you can also blend to make new ones, but the base model does not clone a specific person's voice from a sample. Third-party projects add cloning on top of it; the official release is preset-voice-based.
Use Kompozy to generate the content — the blog, the script, the posts — under your brand voice and to publish it across nine platforms, then run the written script through Kokoro locally to voice a free narration track for a faceless video or a listen-along audio channel. Kompozy owns generation and distribution; Kokoro owns the free, local audio layer.