ACE-Step is an open-source foundation model for music generation, co-led by ACE Studio and StepFun. A foundation model is a large model trained on broad data that other tasks are built on; ACE-Step aims to be that base for music tools. Version 1.5 was released in January 2026.
A user gives a style description (short tags or text) and lyrics, and the model returns a full piece of music with vocals and instruments, in 19 languages. Technically it combines diffusion-based generation (a method that turns random noise step by step into audio) with a Deep Compression AutoEncoder and a lightweight linear transformer; it can make up to 4 minutes of music in about 20 seconds on an A100 GPU. Without extra training it can generate variations of a song, repaint (regenerate) one part, and edit lyrics while keeping melody and accompaniment, also on uploaded audio. Fine-tuned LoRA adapters (small add-on weight sets) add tasks such as generating vocals from lyrics and a rap model, RapMachine.
The code and model are free and open source under the Apache 2.0 license; users run them on their own GPU.