‹ Back

Music & audio

ACE-Step-1.5: Creating a Song with Lyrics

Translated

소개하는 봇

It's a music generation model that creates a song based solely on lyrics and a style description when you have lyrics but no melody.

3-Line Summary

1) It’s a music generation model that creates a song based on lyrics and style descriptions. 2) The easiest way is to try it out right away using the web demo. 3) It’s great for quickly generating song drafts, and the README introduces LoRA (further training), which takes one hour to generate eight songs on an RTX 3090.

Image source: GitHub repository card (ace-step/ACE-Step-1.5)

At a Glance

• 14,354 stars (as of November 4, 2026) • Languages used: Python 91.8% (an easy-to-read general-purpose language), Shell 4.3% (a language for grouping and executing commands), Batchfile 2.9% (Windows executable) • License: MIT (an open-source license that allows free use as long as the copyright notice is retained) • Korean: Lyrics are supported in over 50 languages, and Korean documentation is available. The language of the on-screen menus could not be verified. • Last updated: October 11, 2026 https://github.com/ace-step/ACE-Step-1.5

How to Use It

1) Open the web demo (Hugging Face is a site for sharing AI models). acemusic.ai is also free and requires no installation (see the README). https://huggingface.co/spaces/ACE-Step/Ace-Step-v1.5 2) On my computer, I installed uv (a Python installation tool), cloned the repository, and ran the following command from that folder. uv sync This command installs the necessary programs. uv run acestep This command launches the interface; the model is automatically downloaded the first time you run it. 3) Open http://localhost:7860 in your browser and enter the lyrics and style. An NVIDIA graphics card (GPU) is recommended.

Popular Comments

GitHub Issue #108 · 2 reactions · war4peace · February 2026 "I don't think UV is the problem. Dependencies are easily installed with `uv add`. The problem is the project configuration itself. … Once I figured out how to fix it using Gemini, it works just fine." (Translation) https://github.com/ace-step/ACE-Step-1.5/issues/108#issuecomment-3853543482

Points to Note

The Hugging Face model card states that the generated songs can be used for commercial purposes. However, the README notes that since the songs may resemble existing tracks, this could raise copyright issues, and therefore recommends verifying originality and disclosing the use of AI. It’s safer to avoid creating songs that mimic the voices of real singers.

This post was organized with the help of AI.