What it does ACE-Step UI is a React frontend and Express backend that talks to the ACE-Step 1.5 AI music model via its Gradio API. You type a prompt or paste lyrics, adjust BPM/key/duration, and get back complete songs with vocals—up to 4+ minutes. The UI handles queueing, library management, playlists, and a waveform player, all stored in a local SQLite database.

The interesting bit The project is essentially glue code, but it’s thorough glue: it bundles AudioMass for editing, Demucs for stem separation, FFmpeg for processing, and even a video generator that pulls Pexels backgrounds. The “Thinking Mode” (LLM-assisted generation) auto-disables on GPUs under 12GB VRAM, which is a rare bit of sensible hardware-aware UX in this space.

Key highlights

• One-click start scripts for Windows (start-all.bat) and Linux/macOS (start-all.sh) that launch the Gradio API, backend, and frontend together
• Pinokio integration for genuinely zero-terminal installation
• LAN access baked in—generate from your phone, run the model on your desktop
• Reference audio and “repainting” (regenerating specific sections) for iterative work
• Procedural gradient album art so your library looks finished even offline

Caveats

• Requires ACE-Step 1.5 installed separately; the UI does nothing without it
• NVIDIA GPU required; 4GB VRAM minimum, 12GB+ for full LLM features
• Windows portable package is ~5GB; models download automatically on first run

Verdict Worth a look if you have the hardware and want to own your AI music pipeline end-to-end. Skip it if you’re expecting a hosted service—this is strictly bring-your-own-GPU.