Suno Pro costs $24 a month. Udio costs $30. ACE-Step 1.5 costs nothing — and its makers say it now sits between Suno v4.5 and Suno v5 on quality. It is a free, open-source AI music model that writes and performs complete songs with vocals, in any style, in seconds. It runs on a normal gaming graphics card with less than 4GB of video memory, and if you have no graphics card at all, you can still use it free in your browser. This is the complete guide: what it does, what it needs, and exactly how to make your first song today.
What is ACE-Step 1.5?
ACE-Step 1.5 is an open-source music foundation model built by ACE Studio and StepFun, released in January 2026. The code and model weights are MIT-licensed, the repository has more than 15,000 stars on GitHub, and the model is fast: a full song takes under 2 seconds on an A100 GPU and under 10 seconds on an RTX 3090, according to the project.
It is not a toy demo. It generates complete songs — vocals, instruments, arrangement — from a text description and optional lyrics, from 10 seconds up to 10 minutes long, in more than 50 languages, across 1,000+ instruments and styles. And it does far more than generate:
- Cover generation — remake an existing song in a new style from a reference audio file.
- Repaint and edit — regenerate just one section of a song instead of starting over.
- Track separation — split audio into individual stems.
- Multi-track generation — add layers to a song, like adding instruments one at a time.
- Vocal-to-BGM — automatically generate backing music for a vocal track you already have.
- Fine control — set duration, BPM, key and time signature, or let the model decide.
- LoRA training — train a small add-on model on a handful of your own songs to capture your style (about 8 songs and an hour on an RTX 3090, per the project).
- VST3 plugin — run it inside your music software as a plugin.
Option 1: Try it free in your browser (no install, no GPU)
This is the fastest way, and it costs nothing:
- Open the free demo at huggingface.co/spaces/ACE-Step/Ace-Step-v1.5 — or the project's own free site at acemusic.ai, which the README says is 100% free with no GPU required.
- Type a style description, for example: "upbeat pop, female vocals, acoustic guitar, summer evening mood."
- Add your lyrics (optional) — or leave the lyrics box empty and let the model write them.
- Set the duration, pick a model variant, and click Generate. Download your song as an MP3.
If the Hugging Face Space is busy, use acemusic.ai instead — it is the same model, hosted by the project itself.
Option 2: Run it on your own computer (free, private, unlimited)
Running it locally means no queues, no limits, and your songs never leave your machine. You need Python 3.11 or 3.12 and a GPU is recommended, but the smallest setup runs on under 4GB of VRAM — a mid-range gaming card from years ago is enough. It also supports Mac (Apple Silicon), AMD and Intel GPUs, and even CPU-only mode.
- Install uv, the Python package manager: on macOS or Linux run
curl -LsSf https://astral.sh/uv/install.sh | sh; on Windows run the PowerShell installer from the README. - Clone the repository:
git clone https://github.com/ACE-Step/ACE-Step-1.5.gitand enter the folder:cd ACE-Step-1.5 - Install the dependencies:
uv sync(models download automatically on first run). - Launch the web interface:
uv run acestep— then open http://localhost:7860 in your browser.
Windows and macOS users can also grab a portable package with pre-installed dependencies from the project's download links — see the README's Quick Start section.
The interface picks the best configuration for your GPU automatically. As a rough guide from the README: with 6GB of VRAM or less, use the 2B turbo model with the language model disabled; with 6–8GB, use the 0.6B language model; with 8–16GB, use the 1.7B language model. The larger XL models (4B) need at least 12GB of VRAM with offloading, or 20GB without.
How to get a good song: 5 practical tips
- Be specific about the style. "Lo-fi hip hop, warm Rhodes piano, vinyl crackle, 85 BPM" beats "a nice song." Name the instruments, the mood and the tempo.
- Structure your lyrics. Use [verse], [chorus] and [bridge] tags so the model builds a real song shape instead of one long block.
- Start with the turbo model. It is the fastest variant and ideal for trying ideas. Switch to the sft model when you want the best quality for a final version.
- Use repaint, not regenerate. If one section is weak, select it and repaint just that part — you keep the good sections and save time.
- Generate in batches. The model can make up to 8 songs at once. Generate several versions of the same prompt and keep the best one.
What it costs vs the paid tools
| Tool | Price | What you get |
|---|---|---|
| ACE-Step 1.5 | Free (MIT licence) | Full songs, covers, editing, local or browser, commercial use per the model card |
| Suno Pro | $24/month | Paid generations with monthly credits |
| Udio | $30/month | Paid generations with monthly credits |
ACE-Step's own benchmark claims put it ahead of Suno v5 on SongEval, the standard music-generation benchmark — the project itself describes its quality as between Suno v4.5 and Suno v5. Either way, the price difference is real: the open model is free forever, and you own the setup.
One honest warning
The README itself warns that generated songs can unintentionally resemble existing styles, so check the originality of anything you release commercially and disclose that it was AI-generated. The Hugging Face model card says outputs can be used commercially, and the training data is described as licensed, royalty-free and synthetic music — but the resemblance warning is there for a reason. Treat a generated song like a draft: listen, check, and make it yours before you publish it.
Bottom line
ACE-Step 1.5 is the strongest free answer to paid AI music tools right now: full songs with vocals in seconds, a browser demo that needs no install, and a local version that runs on a modest graphics card. If you make videos, podcasts, games or music, this is a tool worth an hour of your time. Start with the free demo at huggingface.co/spaces/ACE-Step/Ace-Step-v1.5, and if you like what you hear, install it locally from github.com/ace-step/ACE-Step-1.5 and never pay for a song again.
Sources: GitHub — ACE-Step/ACE-Step-1.5 (README, verified live) · Hugging Face — ACE-Step v1.5 demo Space · ACEMusic — free hosted version. Hero photo: a mixing console, CC BY, via Flickr/Openverse.


