Forward Future Tools Library
DiffRhythm
Try DiffRhythm →
aslp-lab.github.io·Open Source
›What is DiffRhythm?
DiffRhythm is a latent diffusion-based song generation model that synthesizes complete songs with both vocal and accompaniment for durations of up to 4m45s in just ten seconds. It is designed to be simple and elegant, eliminating the need for complex data preparation and requiring only lyrics and a style prompt during inference. Its non-autoregressive structure ensures fast inference speeds, making it scalable and efficient for music generation.
›What are DiffRhythm’s key features?
Generates full-length songs with vocals and accompaniment
Fast inference speed of ten seconds for a complete song
Simple model structure requiring only lyrics and style prompt
Scalable and efficient for various music genres
Supports both English and Chinese song generation
›What are the best use cases for DiffRhythm?
Music composition
Songwriting
Content creation
›What is the pricing for DiffRhythm?
Open Source
›Who is DiffRhythm for?
DiffRhythm is built for Musician and Content Creator, working across Design or Music.
