The repository’s news entry is dated April 2, 2026. It is released under the MIT licence.
The project says the XL model can generate audio from 10 seconds to 10 minutes and supports lyrics in more than 50 languages. Listed functions include cover generation, repainting, stem separation, and LoRA training. The repository also lists CUDA, MPS, ROCm, XPU, and CPU as supported execution platforms.
For hardware, ACE-Step says XL models need at least 12GB of VRAM with offload and quantization. A 2B Turbo DiT configuration with CPU offload is presented for GPUs with 6GB or less. The repository claims full-song generation in under two seconds on an A100 and under 10 seconds on an RTX 3090.
The source does not provide an independent quality evaluation, reproducible benchmark methodology, or evidence for its comparison with commercial alternatives.