ACE-Step
An open music-generation model with a Gradio UI for text-to-music, repainting, lyric editing and extending tracks.
Generate songs from tags and lyrics, then retake, repaint, edit lyrics or extend them, on your own GPU.
ACE-Step is an open foundation model for music generation that you run on your own machine through a web interface. It suits technically comfortable musicians and producers who want local generation and editing controls and have a GPU.
What it is
You enter tags and lyrics with structure markers such as [verse] and [chorus], and the model generates a track. The interface also offers Retake, Repainting of selected sections, Edit (lyrics only or remix) and Extend tabs, and the README describes LoRA add-ons such as Lyric2Vocal, Text2Samples and a RapMachine rap LoRA.
Who it's for
It suits producers, songwriters and sound designers who want demo vocals, loops and sketches generated locally. It is not a polished desktop app; you install Python and run it from a terminal.
Key features
Text-to-music tab
Generate tracks from descriptive tags and lyrics with structure markers.
Repainting
Regenerate chosen time ranges of a track while keeping the rest.
Lyric editing
Change small parts of lyrics while preserving melody and accompaniment.
Extend tab
Add music to the start or end of an existing piece.
LoRA applications
Lyric2Vocal and Text2Samples LoRAs make vocal demos and instrument loops.
Low-VRAM mode
CPU offload and overlapped decode options bring maximum VRAM down to about 8GB.
Up and running in 3 steps
Our honest verdict
What's great
- Apache 2.0 licence on the project, with a Gradio interface that exposes retake, repaint, edit and extend.
- The README gives throughput numbers per device, such as about 1.74 seconds to render one minute of audio on an RTX 4090 at 27 steps.
- It supports 19 languages, with the ten strongest named in the README.
- ComfyUI nodes and an API usage path exist for integration into other workflows.
Where it falls short
- It needs Python 3.10 or later, a virtual environment and usually an NVIDIA GPU; a MacBook M2 Max is about 26 seconds per minute at 27 steps per the README.
- Less common languages may underperform because of data imbalance, per the README.
- Lyric edits only work on small segments at a time to avoid distortion.
- Roadmap items such as StemGen and Singing2Accompaniment are listed as coming soon, not shipped.
Pricing & license
ACE-Step is free under the Apache 2.0 licence. The only cost is the GPU or computer you run it on; the README shows no paid hosted plan.
Open remote AI-training roles
768 open remote AI-training roles, updated November 14, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
What hardware do I need?
The README reports speeds on an RTX 4090, A100, RTX 3090 and MacBook M2 Max, and says VRAM can be reduced to about 8GB with CPU offload options.
Can I use the output commercially?
The project is Apache 2.0 licensed and the README does not add a separate weights licence, but it asks you to check originality and disclose AI involvement. Read the full licence before commercial release.
Is there a newer version?
The README news says ACE-Step v1.5 was released on 2026.01.28.
