ACE-Step 1.5

ACE-Step 1.5 is an open-source music generation model co-led by ACE Studio and StepFun, according to its GitHub README. You describe a song in text, optionally add lyrics, and it produces a stereo track between 10 seconds and 10 minutes long. A small language model first plans the song (structure, tempo, key, lyrics) and a diffusion model then generates the audio; the technical report on arXiv, submitted January 31, 2026, describes this split.

Beyond plain generation, the README lists editing tasks: covers of existing audio, regenerating a chosen section (repaint), separating stems, adding layers, and creating backing music for a vocal track. You can also train a LoRA (a small add-on model) on a handful of your own songs to steer its style. It runs through a local Gradio web UI, a REST API, a Python API, a CLI, Hugging Face Diffusers and a separate VST3 plugin for music software. The project says the smallest setup runs in under 4GB of video memory and claims a full song takes under 10 seconds on an RTX 3090; AIGamer hasn't tested it.

For game developers it is a way to make background music and loops offline, then drop the files into any engine. The developer of BIGGER BANG used it for the game's five neighbor-wall tracks. The code and weights are MIT-licensed. The Hugging Face model card says outputs can be used commercially and that training used licensed, royalty-free and synthetic music. Those are the project's own claims.