Is Breeze TTS 2 Free? License and Commercial Use Explained
Breeze TTS 2's weights are free for research and non-commercial use only. Here's what the license actually allows and how to get commercial rights.
Is Breeze TTS 2 free to use?
It depends on what part of the project you’re using and for what. The inference code on GitHub is Apache 2.0, which is genuinely free and open, including for commercial products. The model weights on Hugging Face are a different story: they’re released under the BreezeBlue Research and Non-Commercial License, which restricts use to research and non-commercial purposes. Running the weights, fine-tuning them, or shipping outputs generated by them in a commercial product requires separate written authorization from RESONIA, INC.
TL;DR
- Two separate licenses cover this project: the PyTorch inference code is Apache 2.0, while the model weights fall under a distinct non-commercial license.
- Apache 2.0 does not extend to the model: the model card explicitly states that the code license “does not grant rights to use the model commercially.”
- Weights, checkpoints, adapters, and derivatives are all restricted, meaning fine-tunes or LoRA adapters built on Breeze TTS 2 inherit the same non-commercial terms.
- Self-hosted outputs count too: audio you generate by running the model yourself is covered by the non-commercial license, not just the weights themselves.
- Commercial rights require direct contact with the BreezeBlue team via email, since there’s no self-serve commercial tier described in the documentation.
- Hosted BreezeBlue services, if used instead of self-hosting, run under their own separate service terms, not the research license.
- Users remain responsible for consent and legality around voice cloning and reference audio regardless of which license applies.
What does the Apache 2.0 code license actually cover?
The Apache 2.0 license applies to the inference code published on GitHub, the Python scripts, the streaming API server, the Docker build files, and the general tooling used to run the model. Apache 2.0 is one of the most permissive open-source licenses available. It allows commercial use, modification, redistribution, and even sublicensing of the covered code, with minimal obligations (mainly attribution and preserving license notices).
This matters because it means the engineering around Breeze TTS 2, the way requests are batched, how streaming audio is chunked, the CUDA graph optimizations for fast inference, is not locked behind a restrictive license. A developer could take that code, adapt it, and use it in a commercial pipeline without asking permission, as long as they’re not running it against the restricted model weights for commercial purposes.
The catch is that the code alone doesn’t generate speech. It needs the model checkpoint to do anything useful, and that checkpoint is where the restriction lives.
What does the non-commercial model license restrict?
The BreezeBlue Research and Non-Commercial License governs the model weights, checkpoints, any adapters built on top of them, derivative models trained from them, and the audio output produced by self-hosting the model. The model card is specific about this scope, it’s not just “don’t sell the weights,” it’s a broader restriction covering the whole chain from weights to generated audio.
Practically, this means:
- Downloading and running Breeze TTS 2 locally for personal projects, academic research, or evaluation is permitted.
- Fine-tuning the model on your own voice data and publishing that fine-tune still falls under the non-commercial terms, since derivative models are explicitly named.
- Using the model to generate voice lines for a commercial product, app, game, or client deliverable is not covered by the default license.
- Even audio generated for internal commercial testing would technically fall outside the non-commercial scope, since the restriction applies to self-hosted outputs, not just distribution of the model itself.
This is a common pattern in open-weight releases: the license separates “you can look at this, study it, build research on it” from “you can put this into a paid product.” Breeze TTS 2 draws that line explicitly rather than leaving it ambiguous.
Why release the code as Apache 2.0 but restrict the weights?
Splitting licenses this way lets a lab open its engineering work while still controlling how the trained model itself gets monetized. The code, tokenizers, inference pipeline, and API server represent implementation detail that’s valuable for the community to study, extend, and optimize. Releasing that under Apache 2.0 invites contributions, bug fixes, and integrations without giving away the thing that actually took the compute and data investment to produce, the trained weights.
For BreezeBlue, this structure means outside developers can build faster inference paths, debug the CUDA graph implementation, or wire the API into new frameworks, all of which benefits the ecosystem around the model, while commercial usage of the model itself still routes back through a licensing conversation with the company.
One coffee. One working app.
You bring the idea. Remy manages the project.
How do you get commercial rights to Breeze TTS 2?
According to the model card, commercial use requires written authorization from RESONIA, INC., and the stated path is to contact the BreezeBlue team directly at contact@breeze.blue. There’s no published self-serve commercial license or pricing tier in the documentation, so anyone wanting to embed Breeze TTS 2 in a paid product needs to reach out and negotiate terms individually.
Separately, BreezeBlue also mentions hosted services, meaning you may be able to use the model’s capabilities through a managed API rather than self-hosting the weights. Those hosted services are governed by their own applicable service terms rather than the research and non-commercial license, which gives commercially-minded teams an alternative path that doesn’t necessarily involve running the raw weights yourself.
What counts as “non-commercial” in practice?
The license doesn’t spell out an exhaustive checklist in the transcript-level documentation, but the pattern it follows is standard across similar research licenses: academic research, personal experimentation, open benchmarking, and non-monetized hobby projects are typically fine. Anything tied to revenue, whether that’s a SaaS product, a client project, an app with ads or subscriptions, or an internal business tool, would generally fall under commercial use and need authorization.
Because the license explicitly names “self-hosted outputs” as restricted material, even generating a batch of voice clips for a paying client using your own GPU would fall outside the non-commercial scope, regardless of whether you redistribute the model itself.
Is Breeze TTS 2 worth using if you only need it for research?
For research, evaluation, and non-commercial experimentation, Breeze TTS 2 is straightforwardly free to use. The model card highlights real-time performance characteristics, low time-to-first-audio, and streaming support, along with instruction-following features like voice design and voice direction, all of which make it a reasonable candidate for academic work, benchmarking against other open-weight TTS systems, or prototyping before a commercial licensing conversation. The main thing to keep in mind is that any output or derivative work produced during that research phase stays under the non-commercial terms unless you’ve separately secured commercial rights.
Frequently Asked Questions
Can I use Breeze TTS 2’s code in a commercial product?
Yes, the inference code itself is Apache 2.0 licensed, so you can use, modify, and deploy the code commercially. What you can’t do without separate authorization is run that code against the Breeze TTS 2 model weights for commercial purposes.
Does fine-tuning the model change its license?
No. The non-commercial license explicitly covers “derivative models,” so a fine-tune or adapter built on top of Breeze TTS 2 inherits the same research and non-commercial restrictions as the base weights.
Who do I contact for commercial licensing?
The model card lists contact@breeze.blue as the contact point for commercial licensing inquiries, with written authorization from RESONIA, INC. required before commercial use.
Are outputs generated by self-hosting the model covered by the restriction?
Yes. The license explicitly names self-hosted outputs alongside weights, checkpoints, and derivative models, meaning audio generated by running the model yourself is also restricted to non-commercial use.
Remy doesn't build the plumbing. It inherits it.
Other agents wire up auth, databases, models, and integrations from scratch every time you ask them to build something.
Remy ships with all of it from MindStudio — so every cycle goes into the app you actually want.
Is there a hosted version of Breeze TTS 2 with different terms?
BreezeBlue mentions hosted services governed by their own applicable service terms, separate from the research and non-commercial license that applies to self-hosted weights. Details on those terms would need to come directly from BreezeBlue.
