Sonauto V2.1 is an advanced music generation model developed by Sonauto, designed to transform text prompts, lyrics, or melodies into complete, high-fidelity songs. Built on a latent diffusion architecture, the model handles the entire production pipeline in a single inference, generating vocals, instrumentals, and musical structure across various genres and styles.
The model emphasizes user control and editability, offering features such as inpainting to replace specific audio sections, song extension, and stem separation for isolating individual tracks. It utilizes a comprehensive style tag system and seed-based reproducibility to allow creators to refine musical outputs iteratively. Unlike many competing models that use language-based architectures, Sonauto's diffusion-based approach operates in a continuous space to provide more fluid manipulation of the generated audio.