Sections you can see
The dark gaps in “Windows Down” are where one part hands over to the next. If your take has none, your tags probably didn’t land.
ACE-Step AI music generator
Try the ACE-Step team’s own demo below: tags and lyrics in, up to four minutes of song out.
ACE-Step is an open music model that sings. Its demo takes style tags and lyrics with [verse] and [chorus] markers, and makes 30 to 240 seconds of audio. Its makers publish it under Apache 2.0, and the later v1.5 under MIT.
This is the ACE-Step team’s own demo, on Hugging Face’s free shared machines, so it can queue or stop. It doesn’t use your credits here, and we don’t run it.
Open on Hugging Face- Sings your lyrics
- Up to 240 seconds here
- Retake, repaint, extend
Why section tags matter
ACE-Step reads your lyrics as a map of the song, and sections are what you hear change. This is the loudness of a three-minute ballad from our own generator, for the shape to aim at.
Mark every part
Write [verse], [chorus] or [bridge] on its own line above the words it covers.
No words at all
Put [instrumental] or [inst] in the lyrics box for music with nobody singing.
Genres go in tags
The lyrics box doesn’t take genre tags. Style words go in the tags box, split by commas.
ACE-Step in numbers
From the makers’ pages and this demo.
- Length in this demo
- 30 to 240 seconds; -1 picks one at random
- Length, v1.5
- Up to 10 minutes, says its model page
- Languages
- v1 lists 17 on its page and 19 in its code notes; v1.5 says over 50
- Other tabs
- Retake, repainting, edit and extend
- Published under
- Apache 2.0 for v1, MIT for v1.5
- Cost here
- None: it runs on Hugging Face
What to listen for in your take
Two songs from our generator, drawn as pictures, to show what the tags should be doing.
Fast words need room
Rap like “Late Again” packs in far more words per minute. Give a dense verse a longer duration, or the words get rushed.
ACE-Step’s demo or our song maker
They do a similar job in different ways.
| ACE-Step demo | Our song maker | |
|---|---|---|
| Where it runs | ACE-Step demo: Shared Hugging Face machines | Our song maker: This site |
| What you pay | ACE-Step demo: Nothing | Our song maker: 10 credits for 2 songs on V6 |
| Input | ACE-Step demo: Tags and tagged lyrics | Our song maker: A description, or your lyrics |
| Results kept | ACE-Step demo: Save them from the demo yourself | Our song maker: In your library |
ACE-Step questions
Is ACE-Step free to use?
This demo is free, within whatever Hugging Face allows on shared machines. The model itself is open too, so people with a strong enough computer can run it at home.
Can I use ACE-Step songs commercially?
Its published terms permit it, and the v1.5 page says so outright. Whether your song resembles someone else’s work is a separate question that no model’s terms settle for you.
What’s the difference between ACE-Step v1 and v1.5?
This demo uses v1, capped at 240 seconds. The v1.5 page describes songs of up to 10 minutes and more than 50 languages. We haven’t measured either claim ourselves.
What do retake and repainting do in the demo?
They’re tabs for changing a result you already have: retake makes a variation of the song, and repainting redoes a part of it. There are also tabs to edit and extend.
Not covered here.Ask us something else
Skip the queue and the tags
Describe the song in plain words, and our maker returns two takes to your library.
Make a songTwo free songs, no account needed.