ACE-Step
ace-step/ACE-StepReviewed by AI
Checked by RepoPicks Radar on
What is ACE-Step?
ACE-Step turns a few style tags and your lyrics into a finished piece of music with vocals. Besides writing a new song it can redo a part of one, change the lyrics while keeping the melody, make the track longer or remix audio you upload. It opens as a page in your browser after you install it with Python commands, and it works best with an NVIDIA graphics card with 8 GB of video memory; a MacBook M2 Max also works, more slowly. The model files download by themselves on the first run.
ACE-Step is free and open source (Apache-2.0) and can replace Suno and Udio.
How easy is it to install ACE-Step?
A · download and open · Web
- Windows D
- macOS C
- Linux D
- Web A
The quickest look is the maintainers' hosted "Space Demo" page linked at the top of the README, which needs nothing installed. To run it yourself the README gives only command-line steps: on Mac it is Python, a virtual environment and "pip install -e ." (a C), while Windows and Linux also mean matching PyTorch to your CUDA version for an NVIDIA card, and sometimes "pip install triton-windows", which puts them at D.
ACE-Step on GitHub
- Stars
- 5,045on Nov 10, 2026
- Downloads
- —all releases
- Growth
- +19stars in 6 days
Stars over time
Facts
- Runs on
- Windows, macOS, Linux, Web
- You need to install
- Nothing extra
- To build from source
- Python, Docker, NVIDIA graphics card
- Disk space
- 950 MB–4.3 GB · estimate
- Accounts or keys
- None
- Cost
- Free · Apache-2.0 licence, no account or payment.
- License
- Apache-2.0 (Permissive)
- Last change on GitHub
- Feb 15, 2026
ACE-Step risk signals (4)
- AttentionDependencies with known vulnerabilities
Pinned dependencies with known vulnerabilities (OSV): datasets: GHSA-379c-qx7v-6h59, PYSEC-2026-3716; pytorch_lightning: GHSA-75m9-98v2-hjpm, PYSEC-2026-3043, PYSEC-2026-3967; transformers: GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm, GHSA-489j-g2vx-39wf; accelerate: GHSA-4j2p-28q2-5m79, PYSEC-2026-3804
automatic check
- Attentioncopyright and voice cloning
The project's own disclaimer names "potential risks such as unintentional copyright infringement due to stylistic similarity, inappropriate blending of cultural elements, and misuse for generating harmful content", and the abstract lists "voice cloning, lyric editing, remixing" among the things it enables. Copying a real artist's voice or style can cause legal trouble.
AI review
- Infofake websites warning
The README has an "Important Notice" saying the GitHub Pages site is the only official one, lists masked fake domains ("ac**p.com, a**p.org, a***c.org") and asks you not to "make payments on any of those sites". Only download from the GitHub page.
AI review
- Infodevelopment moved on
The last commit here is from 2026-02-15 and the roadmap still has "Release ControlNet training code" and "Release Singing2Accompaniment ControlNet" unticked, while the news section points to a newer ACE-Step v1.5 released on 2026-01-28 elsewhere. This repository is the older version.
AI review
Evidence is shown in its original language.
What Radar could not check about ACE-Step
- How much disk space the model files really take after the automatic download
- Whether the hosted demo page stays available and how long you wait in its queue
- How slow it is on a computer with no graphics card, since the README only measured NVIDIA cards and a MacBook M2 Max
- Whether the songs sound as good as the examples on the project page
- Whether the install still works today, given the code has not changed since February 2026