deniz.in

Markets

Weather

Loading weather

· via dev.to (home feed)

Open-source AI music model ACE-Step 1.5 claims near-Suno v5 quality for free

ACE Studio and StepFun's MIT-licensed model generates full songs with vocals on modest hardware or in a browser, and its makers place its quality between Suno v4.5 and v5 at zero cost.

Open-source AI music model ACE-Step 1.5 claims near-Suno v5 quality for free

An open-source music generator with a broad feature set

ACE-Step 1.5 is an open-source music foundation model from ACE Studio and StepFun, released in January 2026, according to a hands-on guide published on dev.to. Its code and model weights are MIT-licensed, and its GitHub repository has passed 15,000 stars.

The model writes and performs complete songs — vocals, instruments and arrangement — from a text description plus optional lyrics. Tracks can run from 10 seconds to 10 minutes, span more than 50 languages and draw on over 1,000 instruments and styles. The speed figures come from the project itself: a full song in under two seconds on an A100 GPU, and under 10 seconds on a consumer RTX 3090.

Generation is only part of the package. The dev.to guide lists cover creation from a reference recording, section-level repainting that regenerates one weak passage without discarding the rest, stem separation, multi-track layering, backing music generated for an existing vocal, control over duration, tempo, key and time signature, a VST3 plugin for use inside music software, and LoRA fine-tuning — the project says roughly eight songs and an hour on an RTX 3090 is enough to capture a personal style.

Browser demo or local install

There are two ways to run it, both free. A hosted demo runs on Hugging Face Spaces, and the project also offers its own free site; neither requires a GPU, according to the guide.

Local installation needs Python 3.11 or 3.12 and the uv package manager, after which a web interface launches on the user's own machine. The lightest configuration runs on under 4GB of video memory, and the guide notes support for Apple Silicon, AMD and Intel GPUs as well as CPU-only mode.

Model choice scales with hardware, according to the README guidance cited by dev.to: the 2B turbo variant with the language model disabled fits within 6GB of VRAM, a 0.6B language model suits 6–8GB, a 1.7B model handles 8–16GB, and the 4B XL variants need at least 12GB with offloading, or 20GB without. The interface can also generate up to eight songs in a single batch.

Quality claims versus Suno and Udio

The comparison with commercial tools is the headline. Suno Pro costs $24 a month and Udio $30, while ACE-Step costs nothing — and its licence permits commercial use of the outputs, according to the Hugging Face model card as reported by dev.to.

The quality claims, however, all originate with the project. Its own benchmark results reportedly put it ahead of Suno v5 on SongEval, yet the team simultaneously describes its quality as sitting between Suno v4.5 and Suno v5. Those statements point in slightly different directions, and no independent evaluation is cited; the dev.to article is itself a single write-up drawing on project documentation.

A warning the project itself gives

The README cautions that generated songs can unintentionally resemble existing material, and advises checking originality before commercial release and disclosing that a track was AI-generated. The training data is described as licensed, royalty-free and synthetic music, but the resemblance warning remains, the guide notes.

Why it matters

Music generation has so far been dominated by closed, credit-based subscription services. A capable, MIT-licensed model that produces full songs with vocals, runs on a mid-range gaming card or even a browser tab, and ships with editing tools such as stem separation and repaint changes the economics for hobbyists, video producers, podcasters and game developers. The open question is quality: until independent benchmarks appear, the near-Suno-v5 claim rests on the developers' own numbers. Even so, a zero-cost baseline with commercial-use permissions puts pressure on paid services to justify their monthly fees.

  • #ai-music
  • #open-source
  • #generative-ai
  • #text-to-music
  • #machine-learning

Related posts