ACE-Step 1.5 audio model

In February 2026 ACE-Step 1.5 was released - a free text-to-music model that makes full songs, including vocals and arranged parts. Unlike earlier open models that often sounded lo-fi or experimental, this one aims for polished tracks that could hold up next to commercial stuff.

It’s built for speed and low system needs. That means you can use it for real work like songwriting, mockups, and content creation, not just playing around.

It comes in a few versions. There’s a base model for max control, a fine-tuned one for simpler use, and turbo versions that run faster but might skip a little polish. You can also add optional helper models that guide things like tempo and layout depending on your hardware.

The team behind it is ACE-Step. They’re focused on building open tools for music generation that run on your own machine and don’t hide how they work. The project is still smaller than big-name AI music tools but it’s catching attention for how close it gets to pro sound quality.

Category and cost. It’s a text-to-music tool, totally free under an open license, though you’ll need to check asset rights if you use it commercially.

So what does it do? You give it a prompt and it makes a full track, with vocals and all. You can include layout like intro, verse, chorus, even drops. That gives more control than basic music generators.

Editing is where it really stands out. With repaint, you can tweak just a piece of the track - change the lyrics or fix a bad transition - without starting from scratch. It’s more like using a music editor than just a prompt tool.

It runs fast and light. It works with less than 4GB VRAM and even runs on a CPU (though slower). On a decent GPU it spits out full songs in seconds.

It knows over 50 languages and handles a ton of styles. From metal and K-pop to chiptune, reggaeton, and Indian fusion. Classical music is a weak point though... strings can sound fake.

Here’s what it supports. Text prompts. It turns words into songs. Pick a music style. Choose the kind of sound you want. Adjust instruments. Swap or focus on certain ones. Control the speed. Set the tempo how you like. Set the song layout. Define intros, verses, and more. Make instrumental-only tracks. No vocals if you don’t want 'em. Repaint parts. Just fix one section without redoing the whole thing. Remake songs in new styles. Like a cover but AI-made. Sing in multiple languages. It can even mix them in one track. Keep it private. You can run it all local. Train your own style. With the base model you can fine-tune it.

Team says it performs close to big-name tools like Suno, though take that with a grain of salt.