F5-TTS
A flow-matching text-to-speech model with MIT code, but pre-trained models licensed CC-BY-NC.
MIT code with a Gradio app, CLI and Docker image; pre-trained models are CC-BY-NC.
F5-TTS is a Diffusion Transformer based text-to-speech system trained with flow matching, released with MIT-licensed code. It suits researchers and hobbyists, but its README says the pre-trained models are CC-BY-NC, so they cannot back commercial work.
What it is
You install a PyTorch build for Nvidia, AMD or Intel GPUs, then pip install f5-tts, or run the provided Docker image. It includes a Gradio interface and command-line inference, and can transcribe reference audio with an ASR model if you leave the reference text blank.
Who it's for
It suits researchers, students and hobbyists who want to experiment with flow-matching TTS. It does not suit anyone producing paid or published voice work with the stock models.
Key features
Flow-matching TTS
Diffusion Transformer with ConvNeXt V2 text handling.
Sway Sampling
Inference-time flow step strategy the README says improves performance.
Gradio and CLI
Web interface plus command-line inference.
Docker image
Run with --gpus=all from the published container.
GPU choice
Install notes for Nvidia, AMD and Intel GPUs.
Triton and TensorRT-LLM
A documented runtime deployment for faster serving.
Up and running in 3 steps
Our honest verdict
What's great
- MIT-licensed code you can read and modify freely.
- Install instructions cover Nvidia, AMD and Intel GPUs, not only CUDA.
- A published Docker image gives a fast route to the Gradio interface.
- Leaving the reference text empty makes an ASR model transcribe it for you.
Where it falls short
- The README says pre-trained models are CC-BY-NC because of the Emilia training data, so commercial use is out.
- You need a PyTorch build matched to your GPU, and ROCm 6 breaks on newer AMD cards per the README.
- The first-run model downloads and GPU memory use are yours to manage; no VRAM figure is stated.
- It is a research codebase, not a polished voice editor.
Pricing & license
The code is free under MIT. The pre-trained models are free for non-commercial use only; the only other cost is the GPU hardware you run it on.
Open remote AI-training roles
870 open remote AI-training roles, updated November 14, 2026; most pay $20–$170 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Can I use F5-TTS commercially?
The code is MIT, but the README says the pre-trained models are CC-BY-NC due to the Emilia dataset, so the stock models are not for commercial use.
What GPU does F5-TTS need?
The README has install paths for Nvidia, AMD and Intel GPUs and a CUDA Docker image, but no VRAM figure.
Does it need a reference clip?
The README shows inference with reference audio and optional reference text; leaving the text empty has an ASR model transcribe it.
