
Dia
β Text source-checked Nov 11, 2026Dia is a 1.6B parameter text-to-speech model created by Nari Labs that generates dialogue from a transcript.
Source-checked against the vendor's site Β· rebuilt from the vendor's own pagesProfile updated Link checked How we rate
In short
Dia is a free 1.6B parameter text-to-speech model by Nari Labs that generates realistic English dialogue from transcripts with emotion control and voice cloning.
Dia is a 1.6B parameter text-to-speech model created by Nari Labs. Dia directly generates highly realistic dialogue from a transcript. Users can condition the output on audio to enable emotion and tone control. The model produces nonverbal communications such as laughter, coughing, and clearing throat.
Nari Labs provides access to pretrained model checkpoints and inference code to accelerate research. The model weights are hosted on Hugging Face. A Hugging Face Transformers implementation of Dia is available.
Dia generates dialogue using [S1] and [S2] tags to mark speaker turns. Dia supports voice cloning functionality.
The model currently supports only English generation. The vendor says that Dia has been tested on GPUs with PyTorch 2.0+ and CUDA 12.6. CPU support is planned to be added in the future.
According to Dia: Dia is a 1.6B parameter text-to-speech model created by Nari Labs. Its pages list this strength: The project is licensed under the Apache License 2.0. They also state this limitation: The model supports only English generation.
β The AI Tool Bible editorial team
Pros
- β The project is licensed under the Apache License 2.0.
- β Dia includes a Hugging Face Transformers implementation.
- β Dia supports voice cloning.
- β A ZeroGPU Space is available for trying the model.
Cons
- β οΈ The model supports only English generation.
Facts
- Platforms
- cli (not source-checked)
- Open source
- Yes (license: Apache-2.0)
- GitHub stars
- 19,410 (checked Nov 13, 2026) (not source-checked)
- Last GitHub push
- Nov 19, 2025 (not source-checked)
- First commit
- Apr 2025 (not source-checked)
- Company
- Nari Labs (not source-checked)
- Model used
- Dia-1.6B (not source-checked)
- Hugging Face
- 4 models Β· 35,644 downloads (30 d) Β· 3,289 likes for author nari-labs Β· checked Nov 10, 2026
Where this tool sits
Use cases
Frequently asked
- How much does Dia cost to use?
- Dia is free to use. Nari Labs provides access to pretrained model checkpoints and inference code to accelerate research, with model weights hosted on Hugging Face.
- What languages does Dia support?
- The model currently supports only English generation. The vendor has not documented support for other languages in the provided information.
- Can Dia clone voices?
- Yes, Dia supports voice cloning functionality. Users can also condition the output on audio to enable emotion and tone control in the generated dialogue.
- What hardware is required to run Dia?
- Dia has been tested on GPUs with PyTorch 2.0+ and CUDA 12.6. CPU support is planned to be added in the future, but is not currently available.
- How do I specify different speakers in the transcript?
- Dia generates dialogue using [S1] and [S2] tags to mark speaker turns within the transcript. This allows the model to distinguish between different voices in the conversation.