The infrastructure behind multimodal AI

Nari Labs builds fast, customized and cost-effective multimodal inference. We believe AI-powered multimodal experiences, from speech to world models, will become the default interface of the future. Our goal is to be the infrastructure layer behind that future.

We are experts in multimodal AI research and infrastructure. We built Dia, the first open-source TTS model capable of generating dialogue. The Dia series has been downloaded more than 2 million times, reached #1 on Hugging Face, and earned more than 20,000 GitHub stars. We also built an OSS TTS inference engine that achieves SOTA latency and reached #1 on benchmarks.

Our ex-YC and ex-KRAFTON team of engineers has published at NeurIPS and ICLR, and earned three gold medals across the IOI and ICPC World Finals. Nari Labs is backed by Y Combinator. We are always looking for exceptional people to join our team: Contact Us.