ACE-Step-1.5
ace-step/ACE-Step-1.5 · Python · MIT
Local full-song music generation that fits on a consumer GPU
ACE-Step 1.5 aims at commercial-grade text-to-music with lyrics on machines many creators already own. It supports covers, remix, LoRA personalization and newer XL variants when you have more VRAM.
Score
Setup at a glance
git clone https://github.com/ace-step/ACE-Step-1.5.git && pip install -r requirements.txt Deep dive
ACE-Step is an open-source music foundation model: give it lyrics and a style prompt and it generates full songs — vocals, arrangement and mix — on consumer GPUs, without a studio or an API bill.
The 1.5 generation adds the XL series (4B DiT) for higher quality, plus LoRA personalization so you can fine-tune the model toward your own voice or style, and dedicated cover and remix modes for reworking existing tracks.
It ships with launch scripts for common setups, a Windows one-click download, and base/SFT/turbo checkpoints on Hugging Face so you can trade quality against speed.