F5-TTS

What does F5-TTS do?

F5-TTS is a text-to-speech model from a research paper, described in its README as a "Diffusion Transformer with ConvNeXt V2, faster trained and inference". It clones a voice from a reference recording and its transcript, with no training step. The bundled Gradio web app offers basic generation with chunked inference, multi-style and multi-speaker generation, and a voice chat demo. Training and fine-tuning are documented, with a Gradio app for the fine-tuning as well.

The released checkpoints were trained on the Emilia dataset in Chinese and English. A separate document covers custom inference in other languages. Streaming, model size and sample rate are not stated in the README; the vocoder it names runs at 24 kHz.

Key facts

• Licence on the code: MIT. Licence on the pretrained weights: CC-BY-NC-4.0, stated in the README and on the Hugging Face card at SWivid/F5-TTS.
• The README's own words: "The pre-trained models are licensed under the CC-BY-NC license due to the training data Emilia, which is an in-the-wild dataset."
• Zero-shot cloning from a reference audio file and its transcript; leaving the transcript empty makes the tool transcribe the clip itself at extra GPU memory cost.
• Released checkpoints: F5TTS_v1_Base, F5TTS_Base and E2TTS_Base, trained on Emilia in Chinese and English.
• The Gradio app offers basic generation with chunked inference, multi-style and multi-speaker generation, and a voice chat demo.
• Install paths for NVIDIA GPUs, AMD GPUs on ROCm (Linux only), Intel GPUs and Apple Silicon; a Docker image at ghcr.io/swivid/f5-tts.
• A Triton and TensorRT-LLM runtime is documented, with the project's figure of a 0.0394 real-time factor on one L20 GPU.
• Latest tagged release when read: 1.1.22 on 2026-07-23.

What does it replace, and where does it fall short?

F5-TTS replaces ElevenLabs for research and personal projects that need a voice clone, and for teams that intend to train their own weights on data they have the rights to.

It does not replace ElevenLabs for a commercial product as shipped, and this is the reason for the open-weights label. The README says: "The pre-trained models are licensed under the CC-BY-NC license due to the training data Emilia, which is an in-the-wild dataset." The code is MIT. So you may use, change and sell software built on the code, and you may not use the published weights commercially.

How does F5-TTS run?

pip install f5-tts, or clone and install in editable mode, or the Docker image at ghcr.io/swivid/f5-tts. There is a command-line tool and a Gradio web app. The README gives PyTorch install paths for NVIDIA, AMD (ROCm, Linux only), Intel GPUs and Apple Silicon. A Triton and TensorRT-LLM runtime is documented, with the project's own figure of a 0.0394 real-time factor on one L20 GPU. Nobody sells a hosted version; a third-party demo on Hugging Face is linked.

Who is F5-TTS for?

A researcher, or a team that will train its own checkpoint on licensed audio and wants a working, permissively licensed pipeline to do it in. A team that wants to ship cloned voices under a clear commercial licence should use Chatterbox.

What limits does the README state?

Leaving the reference transcript empty makes the tool transcribe the clip itself, "extra GPU memory usage". Some AMD GPUs fail on ROCm 6.x with an invalid device function error. The maintainers write, of the non-commercial weights, "Sorry for any inconvenience this may cause."

Questions people ask

Is F5-TTS open source?

The code is: MIT, an OSI approved licence. The published weights are not: CC-BY-NC-4.0 forbids commercial use. This site labels a project open weights when the weights carry a restriction the code does not, and F5-TTS is the clearest example of that shape.

Can F5-TTS be used commercially?

The code, yes, under MIT. The published weights, no: CC-BY-NC-4.0 excludes commercial use, and the README apologises for the inconvenience. A team that trains its own checkpoint on audio it has the rights to, using the documented training pipeline, owns the result under the terms of its own data.

Does F5-TTS run offline, and what hardware does it need?

Yes, on your own machine. The README documents PyTorch installs for NVIDIA, AMD ROCm on Linux, Intel GPUs and Apple Silicon; CPU-only use is not described. Some AMD GPUs fail on ROCm 6.x with an invalid device function error, which the README notes.

How does F5-TTS compare with ElevenLabs?

For a demo or a research project, F5-TTS clones a voice from one clip on your own GPU with no bill. For a product, its weights cannot be used, where ElevenLabs' terms and Chatterbox's MIT weights both can. The open-source alternatives to ElevenLabs page places it as the research and training pick.

Can I fine-tune F5-TTS on my own voice data?

Yes. The README documents training with Accelerate and a Gradio app for fine-tuning, and links a guide for custom inference in more languages. The released checkpoints cover Chinese and English; other languages need your own training or a community checkpoint, each with its own licence to read.

Sources

• F5-TTS README and MIT LICENSE: github.com/SWivid/F5-TTS, read 2026-09-16.
• Model card, cc-by-nc-4.0: huggingface.co/SWivid/F5-TTS, read 2026-09-16.