ACE-Step Studio

An AI music generator for Windows that runs ACE-Step locally on NVIDIA GPUs, with vocal songs, covers, stem separation and LoRA training.

ACE-Step Studio is a desktop AI music generator for musicians and anyone who wants to create songs on their own hardware. It turns a short idea, or a description with lyrics, into a full vocal track or instrumental.

Song creation and editing: Generate songs with ACE-Step's planner, reinterpret an existing track in another style, make remixes or regenerate a selected passage.
Style adaptation: Browse LoRAs on Hugging Face or train your own from songs in a particular style using the local HOT-Step trainer.
Stems, MIDI and karaoke: Separate audio with HT-Demucs, convert recordings to multi-instrument MIDI with MuScriptor, and generate lyric timings locally with Parakeet or Whisper.
Audio finishing: Apply noise reduction, use VST3 plugins and master audio against a reference track.
Agent control: Connect Claude Code, Claude Desktop or Cursor through MCP so an AI agent can write songs and operate the studio.

The free, open-source app uses the MIT license and runs on Windows 10/11 x64. Its supported GPU setup starts with an NVIDIA GTX 900-series card and 4 GB of VRAM. It uses ACE-Step models in GGUF format and works offline after model downloads, without an account or telemetry. Music stays on your disk unless you enable a cloud capability.

The optional writing assistant can use local Gemma or OpenRouter; AI-generated cover art uses OpenRouter. LoRA training requires an NVIDIA RTX card, while audio-to-MIDI requires a GTX 16-series or RTX 20-series card or newer. The Regrind fine-tune uses CC BY-NC-SA 4.0, and catalogue LoRAs carry their authors' licenses.