Free and MIT-licensed, so musicians and developers can run or fine-tune it on modest hardware, with an online demo for those who can't.
Kenning's take
What it does
ACE-Step 1.5 is an open-source model that generates songs from a text description and lyrics, from short loops up to ten minutes. Its makers say it produces a full song in under 2 seconds on an A100 and under 10 seconds on an RTX 3090, needs under 4GB of video memory, and supports covers, editing parts of a track and training a personal style from a few songs. It is free to download from GitHub and Hugging Face under the MIT licence, and can be tried online at acemusic.ai.
- Songs from 10 seconds to 10 minutes
- Lyrics in 50+ languages
- Cover generation, repaint and edit
- Vocal-to-accompaniment and track separation
- LoRA style training from a few songs
- Runs locally with under 4GB VRAM
Who it suits. Musicians, producers and developers who want a free music model they can run or fine-tune on their own hardware.
Key facts
- Category
- AI models
- Made by
- ACE Studio and StepFun
- Released
- 31 January 2026
- How to get it
- free download of the weights and code from GitHub and Hugging Face (MIT licence); free online demo at acemusic.ai and a Hugging Face Space
- Works on
- Windows, macOS, Linux, Web
- In the UK
- Yes
- Free plan
- fully free: open weights under MIT licence
FAQ
When was ACE-Step 1.5 released?
The technical report was posted on 31 January 2026. A larger XL version followed on 2 April 2026.
Is ACE-Step 1.5 free?
Yes. The code and weights are open source under the MIT licence, and the acemusic.ai site is described as free.
What hardware does ACE-Step 1.5 need?
The makers say it runs locally with under 4GB of video memory; the XL version needs at least 12GB.
Facts from the maker's own pages and the sources listed. Checked 2026-10-11.
Similar tools
AlphaGenome Atlas
Google DeepMind database of AI-predicted effects for every possible single-letter change in human DNA.
Amazon Nova 2
Amazon's second family of Nova AI models, covering reasoning, speech conversation and combined text-and-image generation.
Cisco Antares
Cisco's small open-weight models that find where known security vulnerabilities sit in a codebase, running locally.
QuiverAI Arrow 2
QuiverAI's model for generating, vectorising and animating SVG graphics, released with a higher-end Telos version.