The Patchbay_

YuE2 vs ACE-Step

The only two open models that turn lyrics into full songs with vocals. YuE2 sounds better; ACE-Step is the one you can actually ship and actually run.

Short answer. Shipping anything, or on modest hardware or a Mac → ACE-Step. Chasing the best open song quality, with Linux and 24 GB → YuE2.

Side by side

YuE2 ACE-Step
Audio qualityBetter — competitive with Suno v5Good, clearly behind YuE2
Licence (weights)CC BY-NC 4.0 — non-commercialApache-2.0
Output48 kHz stereoHigh quality
VRAM24 GB, no quantised fallback8 GB with offload flags
OSLinux onlyLinux, Windows, macOS
Apple SiliconNoYes — ~1× real time on M2 Max
Editable scoreYes — ABC notationNo
EditingEdit score, re-renderRetake, repaint, edit, extend
CoversZero-shot via SheetSage2audio-to-audio
LanguagePythonPython
LicenseApache-2.0 (code) · CC BY-NC 4.0 (weights)Apache-2.0
PlatformsLinuxLinux Windows macOS
First released2025 (YuE2 supersedes it)2025
MaintainedYesYes

Where they actually differ

Licensing points one way

ACE-Step is Apache-2.0 on code and weights. YuE2’s code is Apache-2.0 but its weights are CC BY-NC 4.0 — non-commercial. Older write-ups describing YuE as the permissive open answer to Suno are describing version 1 and are out of date. For any commercial use, ACE-Step is the only option of the two.

Hardware points the same way

YuE2 documents Linux, Python 3.12 and a BF16-capable 24 GB NVIDIA GPU, with no quantised path. ACE-Step runs in 8 GB with offload flags and is the one serious song model that works properly on Apple Silicon. Between them, most people can only run one.

Quality points the other way

On the project’s own WildSongBench evaluation, YuE2 scores 6.73 (6.96 best-of-8) against ACE-Step 1.5 at 6.01, with YuE2 sitting alongside Suno v5. Treat authors’ own benchmarks carefully, but the gap is large enough to be real.

They edit very differently

YuE2 exposes the composition as ABC notation you can change before rendering — genuinely unusual. ACE-Step edits the audio result: retake with variance, repaint a time range, change lyrics while preserving melody, extend. One is composing, the other is producing.

Which should you choose?

Choose YuE2 when…

Choose ACE-Step when…

YuE2 detailsACE-Step detailsAll ML & generative

← All comparisons