Can you ship a game made with Kokoro-82M? A pricing page names a commercial license. The terms quoted beside it may narrow that. The model card says Kokoro is Apache-licensed, has been deployed in commercial APIs, and the author welcomes real-world deployments.
Kokoro-82M is an open-weight text-to-speech model with 82 million parameters and Apache-licensed weights, so you can run it locally for free. Its model card says it was trained only on permissive or non-copyrighted audio. Its usage examples install espeak-ng, which is GPL v3 or later.
A pricing page names a commercial license. The terms quoted beside it may narrow that.
The model card says Kokoro was trained only on permissive or non-copyrighted audio and IPA phoneme labels.
Kokoro's README installs espeak-ng for English out-of-dictionary fallback and some non-English languages. espeak-ng is released under GPL v3 or later, so check how your build bundles it.
Voice lines you generate during development and ship are Pre-Generated AI content on Steam. If the game synthesizes speech at runtime, the Live-Generated rules apply too.
Quick answers
Can I use Kokoro-82M commercially in a game?
The model card says Kokoro is Apache-licensed, has been deployed in commercial APIs, and the author welcomes real-world deployments.
How much does Kokoro-82M cost?
The weights are free to download and run. The model card puts the April 2025 market rate for Kokoro served over third-party APIs at under $1 per million characters (under $0.06 per hour of audio).
Built only from rows with a stored vendor quote. The facts below carry the exact sentence, source link, and fetch date.
Facts
Verified means the quote was copied from the source page on the fetched date. It goes stale after 30 days until the checker sees it again.
Free weights; hosted APIs from third parties
PricingVerified
The weights are free to download and run. The model card puts the April 2025 market rate for Kokoro served over third-party APIs at under $1 per million characters (under $0.06 per hour of audio).
As of April 2025, the market rate of Kokoro served over API is under $1 per million characters of text input , or under $0.06 per hour of audio output.
The model card says Kokoro is Apache-licensed, has been deployed in commercial APIs, and the author welcomes real-world deployments.
This is an Apache-licensed model, and Kokoro has been deployed in numerous projects and commercial APIs. We welcome the deployment of the model in real use cases.
Kokoro's README installs espeak-ng for English out-of-dictionary fallback and some non-English languages. espeak-ng is released under GPL v3 or later, so check how your build bundles it.
eSpeak NG Text-to-Speech is released under the GPL version 3 or later license.
Voice lines you generate during development and ship are Pre-Generated AI content on Steam. If the game synthesizes speech at runtime, the Live-Generated rules apply too.
Pre-Generated: Any kind of content that ships with your game and is consumed by players that is created with the help of AI tools during development.
Altered Studio — Voice changer, text-to-speech, and voice cloning app (Altered Ltd), usable for temp VO and character voices. Licensing is set by plan: Free is attribution-only and non-commercial, Creator allows commercial use under $100k revenue or funding, and Professional carries the full commercial license. Altered keeps ownership of its stock Voices.
Cartesia — Cartesia sells text-to-speech, speech-to-text, and voice agents. Pro is the first card that lists a commercial use license. Output ownership was not on the pricing page.
ElevenLabs — Speech, sound effects, and music generation. The terms allow commercial use on paid plans and restrict free use to non-commercial. Starter is listed at $6 per month, with a $1 first-month promo shown until Nov 1, 2026.
Fish Audio — AI text-to-speech and voice cloning from Hanabi AI. Free tier at $0 (non-commercial), Plus $15/month ($11 billed annually), Pro $100/month. The terms license paid users for commercial use and allow training on your content; output ownership is not stated.
Piper — Piper is a free, open-source text-to-speech engine that runs locally on CPU, with pre-trained voices in dozens of languages. The engine is GPL-3.0-or-later, and every voice carries its own license in its MODEL_CARD: some are CC0 or CC BY, others are non-commercial or research-only.
Typecast — Typecast, from Neosapience, is an AI text-to-speech and voice-cloning studio with a library of character and game voices for NPC dialogue, narration, and cutscenes. Generating and playing back audio is free on every plan; credits are spent when you download. Free gives 3,000 lifetime credits and requires attribution; Basic is $5 a month and adds a commercial license, then Plus $19, Pro $29, and Business $69 (businesses, institutions, and agencies are told to use Business). The terms say you own the output during your subscription, and you can keep using audio you downloaded after it ends. Selling voices as standalone files or sound libraries is not allowed, and API or automated integrations need a separate agreement.
WellSaid — WellSaid is an AI voiceover studio with hundreds of voices. The free trial says it has no commercial rights; Starter and Pro list full commercial rights, metered by downloaded minutes. Output ownership was not on the pricing page.