ACE-Step 1.5

1. What is ACE-Step 1.5?

ACE-Step 1.5 is an open-source AI music model. You describe the song you want (for example "upbeat electronic dance music with heavy bass") or paste lyrics, press a button, and it generates a full track with instruments and, optionally, vocals. It runs on your own computer, and the interface is a web page you open in your browser. The project compares its quality with commercial music services such as Suno; that comparison comes from the authors, so listen for yourself.

It is free, MIT-licensed and needs no account. Running it on your own machine has no per-song fee.

A good fit if:

• You need background music for videos, podcasts or games and want to try many versions without an online service's generation limits.
• Your PC has an NVIDIA graphics card (the documentation says about 4 GB of card memory is enough for the basic mode).
• You are fine with a few setup steps and one large download.

Not a good fit if:

• You have an office laptop without a dedicated graphics card. The documentation says the CPU works but is clearly slower. In that case try the web demo the project points to (see section 7).
• You need Vietnamese vocals to be reliable. The README says lyrics work in 50+ languages but does not list them by name, so test Vietnamese yourself first.
• You plan to monetise the music or publish it where copyright is checked. The authors advise checking that your work is original and disclosing AI use. That is your responsibility; the project accepts none.

2. What you need

According to the official documentation:

	Requirement
Operating system	Windows (this guide focuses on Windows); macOS and Linux are also supported
Graphics card	An NVIDIA card with CUDA is recommended. The ready-made Windows package requires CUDA 12.8
Card memory (VRAM)	4 GB for the basic mode (music only); 6 GB if you also enable the language model that helps write lyrics and refine songs
Free disk space	About 10 GB for the core models, plus room for the package
Python	With the Windows "portable" package you do not install Python yourself; it is included

The "portable" package has everything pre-installed: you just unpack and run it. It is about 2.4 GB. To unpack the .7z file you need 7-Zip (free, https://www.7-zip.org).

On the first run the program downloads more AI models from the internet, so you need a stable connection, and a slow one can mean waiting a while.

The latest GitHub release when this was written is v0.1.8 (18 May 2026); the code has been updated since.

3. Step-by-step installation

Step 1: Download the Windows portable package

Open https://github.com/ACE-Step/ACE-Step-1.5 and use the "portable package" link in the README's Quick Start, or open it directly:

https://files.acemusic.ai/acemusic/win/ACE-Step-1.5.7z

Safety note: the authors warn that fake domains imitate the project. Only use links that lead from the official GitHub page above, not ads or random posts.

What you should see: a file ACE-Step-1.5.7z of about 2.4 GB.

Step 2: Unpack it

Right-click the file and choose 7-Zip → Extract to "ACE-Step-1.5". Pick a drive with plenty of free space and a short path without accented characters, for example D:\ACE-Step-1.5.

What you should see: the folder contains files such as start_gradio_ui.bat and a python_embedded folder.

Step 3: Start the interface

Double-click start_gradio_ui.bat.

What you should see: a black command window opens and prints text. On the first run it downloads models, so it may sit still for a long time. Do not close this window while you use the program. When it is ready, open your browser at:

http://localhost:7860

You will see the ACE-Step web interface (a Gradio page with sections such as Settings, Generation and Results).

Alternative (if you are comfortable with the command line): install with uv

You need Python 3.11 or 3.12 and Git. The documentation installs the uv tool and then runs:

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
git clone https://github.com/ACE-Step/ACE-Step-1.5.git
cd ACE-Step-1.5
uv sync
uv run acestep

Then open http://localhost:7860 as well.

4. First use

1. In the web interface, open Settings and click Initialize Service to load the models. The status box reports the graphics card and memory it detected, and adjusts the maximum duration to suit your machine.
2. In the Generation tab, choose the Simple mode.
3. Type a description in "Song Description". The documentation's example: "jazz trio playing in a smoky bar". For music without vocals, tick Instrumental.
4. Click Create Sample so the program writes a caption and lyrics, then click Generate Music.
5. In the results area, press play to listen and download the file.

What you should see: after a few minutes (much faster with a strong card) one or more tracks with play buttons. The number of tracks per run (Batch Size) defaults to 2.

To write your own lyrics, use the Custom mode and divide them with tags such as [Verse 1] and [Chorus].

5. Common problems and fixes

Sign	Cause	Fix
Window says "No GPU detected, running on CPU"	The graphics card was not recognised, or a CPU-only PyTorch build is installed	Update the NVIDIA driver; run the project's diagnostic python scripts/check_gpu.py; for AMD cards use start_gradio_ui_rocm.bat
Out-of-memory error	The card has too little VRAM for the chosen model	Use a smaller model (the documentation suggests "2B turbo" for cards under 8 GB), turn off the language model, shorten the duration or generate fewer tracks at once
The browser cannot open http://localhost:7860	The program has not finished starting, or port 7860 is used by another app	Wait until the black window says it is ready; close the other app using the port; the port can be changed in the configuration (PORT)
Stuck for a long time downloading models	Large models and a slow or blocked connection	Wait; try changing the download source in the .bat file (DOWNLOAD_SOURCE, with huggingface or modelscope)
Windows or antivirus blocks the .bat file	Unsigned script	Allow it only if you downloaded from the official source in Step 1

6. Uninstalling and updating

Uninstall: the portable package does not install into the system, so close the black window and delete the ACE-Step-1.5 folder you unpacked. Downloaded models may live in this folder or in the download tool's cache; the documentation does not say, so after deleting check your disk to confirm the space came back.

Update: the package includes check_update.bat, which checks for and applies updates from GitHub. The documentation says automatic updates need a PortableGit folder inside the package. If that sounds complicated, download the newer portable package and unpack it into a different folder.

7. Frequently asked questions

Does it cost money? No. The project is free under the MIT licence.

What if I do not have an NVIDIA card? The documentation says it can run on the CPU, on Apple Silicon Macs, or on AMD and Intel cards, but slower or with a different setup. The README also points to a free demo that needs no graphics card at https://acemusic.ai ; that site is named by the project, but weigh it yourself before using any web service.

Is my data sent anywhere? When you run it on your own machine as in section 3, the models run locally. Only the first run downloads models from the internet.

Can I use the music commercially? The README only says the code is MIT-licensed and advises checking for stylistic similarity and disclosing AI use. For commercial work, do your own research.

Is the interface available in Vietnamese? In the portable package's configuration, the interface-language line lists en, zh, he and ja. Vietnamese is not among them.