ace-step src ddc969cb3d3396f64cc27de57321a15c0b66af435725272bff4ace434cfd328d Local text-to-music engine (ACE-Step 1.5) ACE-Step 1.5 generates full songs (instrumental backing plus optional vocals from lyrics) on a local GPU. This package ships the engine, a local OpenRouter-compatible API on 127.0.0.1:8002, and a config-driven CLI. The Gradio web UI is not packaged yet. Checkpoints are not included. Download them once with: acestep --download The wrapper looks for weights in ACESTEP_CHECKPOINTS_DIR, then /media/space/ai/ace-step, /srv/ai/ace-step, and finally $XDG_DATA_HOME/ace-step/checkpoints. acestep # local API (ComfyUI talks to this) acestep --cli # interactive / config CLI acestep --download # fetch checkpoints from Hugging Face systemctl --user enable --now ace-step The 5Hz language-model backend defaults to PyTorch (pt). That is the supported path on ROCm; nano-vllm is not shipped. bero <bero@lindev.ch> https://github.com/ace-step/ACE-Step-1.5