ACE-Step 1.5 XL lands with 4B models
ACE-Step 1.5 XL is the new 4B music-generation tier from the ACE-Step project, released as base, SFT, and turbo variants on Hugging Face. The release pushes an open-source, MIT-licensed text-to-music stack aimed at commercial-ready use, with the base and SFT models emphasizing quality and prompt adherence, and the turbo model targeting faster 8-step inference. The project says the XL models work with the existing ACE-Step LM stack, support local runs with modest VRAM via offload/quantization, and cover editing and composition use cases beyond simple text-to-music.
Hot take: this is one of the more serious open music model drops in a while, but it still looks like a power-user tool more than a push-button consumer app.
- –The lineup is clear: `base` for broad task coverage, `sft` for the best quality and CFG control, and `turbo` for speed.
- –The project claims commercial-ready training data and MIT licensing, which makes it easier to actually use than many music-gen releases.
- –The main tradeoff is prompt discipline: the model appears capable, but it wants detailed, structured instructions to shine.
- –For local users, the hardware story matters as much as the model quality; the release is positioned around consumer GPUs rather than only datacenter rigs.
DISCOVERED
52d ago
2026-04-07
PUBLISHED
52d ago
2026-04-07
RELEVANCE
AUTHOR
Uncle___Marty
