Dia nari-labs/dia

1.6B-parameter English text-to-dialogue model from Nari Labs with two-speaker tags, non-verbal sounds and audio-prompt voice cloning.

Creator kit + Compare GitHub ↗ Install guide (official) ↗

Dia at a glance

Checked 16 Oct 2026

Dia generates two-speaker dialogue ([S1]/[S2] tags) with non-verbals like laughter, and can clone a voice from a 5-10 second audio prompt. Apache-2.0 code and Apache-2.0 weights on Hugging Face. English only. Tested on GPU only (about 4.4 GB VRAM in bfloat16 on an RTX 4090); CPU support is listed as coming. The README describes it as intended for research and educational use. A successor, Dia2, lives in a separate repo (nari-labs/dia2).

Cite this pageRepoQuarry. “Dia.” Checked 16 Oct 2026. https://repoquarry.dev/projects/dia

How much of ElevenLabs does it replace?

2 of 4 jobs. English only, dialogue-oriented; voices vary per run unless you use an audio prompt or fix the seed.

Speech generation Partial replacement

Voice cloning Partial replacement

Speech recognition Not covered

Voice agents Not covered

For speech recognition, voice agents, see other ElevenLabs alternatives. You may also need to build an interface, an API server and hosting yourself.

How much of Descript does it replace?

2 of 3 jobs. Voice cloning and text-to-speech only; no transcript-based audio or video editing.

Transcript-based editing Not covered

Voice cloning Partial replacement

Speech generation Partial replacement

For transcript-based editing, see other Descript alternatives. You may also need to build an interface, an API server and hosting yourself.

Facts, with sources

Every answer links to where we found it. We read the documentation; we did not install it.

Report a correction

QuestionAnswerSourceChecked

Runs on my machine? Yes, on an NVIDIA GPU (CPU support not yet available) Dia has been tested on only GPUs (pytorch 2.0+, CUDA 12.6). CPU support is to be added soon. github.com · dia#-hardware-and-inference-speed ↗ 1d ago

Code licence Apache-2.0 This project is licensed under the Apache License 2.0 - see the LICENSE file for details. github.com · LICENSE ↗ 1d ago

Model weights licence Apache-2.0 (Dia-1.6B and Dia-1.6B-0626 on Hugging Face) license: apache-2.0 huggingface.co · Dia-1.6B-0626 ↗ 1d ago

Commercial use Yes by licence (Apache-2.0 code and weights), though the README says the model is intended for research and educational use and forbids identity misuse and deceptive content This project offers a high-fidelity speech generation model intended for research and educational use. github.com · dia#%EF%B8%8F-disclaimer ↗ 1d ago

Paid services needed? None required The model weights are hosted on Hugging Face. github.com · dia#readme ↗ 1d ago

Hardware NVIDIA GPU; about 4.4 GB VRAM in bfloat16/float16, about 7.9 GB in float32 (benchmarked on RTX 4090) These are the speed we benchmarked in RTX 4090. github.com · dia#-hardware-and-inference-speed ↗ 1d ago

Operating systems README lists Docker support for ARM and macOS as future work; it does not provide a verified native Windows/macOS/Linux support matrix Docker support for ARM architecture and MacOS. github.com · dia#readme ↗ today

Setup effort Easy, pip install from GitHub (or uv), or via Hugging Face Transformers pip install git+https://github.com/nari-labs/dia.git github.com · dia#quickstart ↗ 1d ago