Which local TTS engines allow commercial use?
Qwen3-TTS (0.6B and 1.7B), VoxCPM2, Audio8 TTS Preview and NeuTTS Air are Apache 2.0, and Chatterbox Turbo is MIT: all allow commercial use. Breeze TTS 2 and Fish S2 Pro are non-commercial. KittenTTS 2 is free for small businesses under US$1M, with registration.
License by engine
| Engine | License | Commercial use |
|---|---|---|
| Qwen3-TTS 0.6B | Apache 2.0 | Yes |
| Qwen3-TTS 1.7B | Apache 2.0 | Yes |
| Chatterbox Turbo | MIT | Yes. Adds Resemble AI's inaudible watermark to everything it makes |
| VoxCPM2 | Apache 2.0 | Yes |
| Audio8 TTS Preview 0.6B | Apache 2.0 | Yes |
| NeuTTS Air | Apache 2.0 | Yes |
| KittenTTS 2 | Stellon Labs Community | Free for personal use, research and small businesses under US$1M revenue and funding; commercial users must register with Stellon Labs; outputs may not be used to train other models |
| Breeze TTS 2 | Non-commercial | No |
| Fish S2 Pro | See its model card | No, non-commercial |
This is a summary of the licenses as Mimic's documentation lists them, not legal advice. Read the license on each model's page before using its output in a product, and recheck when a new version comes out.
Licenses aren't the only rule
Whatever the license, you need permission from the person whose voice you clone, and many places have laws on impersonation, robocalls and deepfakes. Mimic asks you to confirm consent before your first clone and marks saved takes as AI-generated in the file's metadata. When you share synthetic speech, say so.
Practical picks
- Commercial and light on hardware: Qwen3-TTS 0.6B, at 2.4 GB of VRAM or 5.6 GB of RAM. See 0.6B vs 1.7B.
- Commercial with only a clip, no transcript: Chatterbox Turbo, from a clip over 5 seconds, at 3.4 GB of VRAM.
- No GPU: see voice cloning without a GPU.