DiffRhythm is a software that transforms lyrics into complete songs by generating professional-quality music with vocals from your lyrics in less than 15 seconds. It allows creating unlimited songs in various styles and languages.
Features
Blazing Fast Generation
Generate full-length songs up to 4m45s in just ten seconds, dramatically faster than other music generation systems.
Complete Song Generation
Produces both vocal and accompaniment tracks simultaneously in a single process, maintaining high musicality and intelligibility.
Multi-Language Support
Create songs in multiple languages including English and Chinese with natural pronunciation.
Embarrassingly Simple Design
Straightforward model structure eliminates the need for complex data preparation or multi-stage cascading architectures.
Style Prompt Control
Requires only lyrics and a style prompt during inference to generate diverse musical outputs across different genres.
Latent Diffusion Model
First latent diffusion-based song generation model, moving beyond slower language model-based approaches.
Non-Autoregressive Structure
Architecture ensures fast inference speeds compared to autoregressive models that generate content sequentially.
Highly Scalable
Simple yet powerful design guarantees scalability for future development and broader applications.
Open Source
Available on GitHub and Hugging Face with demo examples, making it accessible for researchers and developers.