ACE-Step 1.5 tutorial: music generation and stem editing

This tutorial covers ACE-Step 1.5, an open source music model, through its local Gradio interface. The presenter explains its language model planner and diffusion audio engine, then demonstrates song generation and editing. His computer has 16 GB VRAM and 64 GB RAM; these are demonstration specs, not stated minimum requirements.

Text-to-music gets the most favorable assessment. The walkthrough covers structured lyrics, music captions, BPM, key, fixed seeds and instrumental generation. The presenter uses eight steps with Turbo and describes roughly 32 steps for Base. He reports generation times of 15 to 30 seconds on his setup and switches to Base for extraction, layering and completion.

The editing examples show uneven results. Repainting mutes a selected passage in one attempt, and a later lyric replacement still sounds unclear. For covers, the presenter finds that lowering audio cover strength to around 0.2 improves his acoustic conversion. Vocal and drum extraction fare better in his tests than bass or keyboard extraction.

Layering and completion produce results he considers experimental, including silent or missing outputs. He also reports that disabling Think restores audio in some workflows. These observations describe his tests, rather than guarantees about every installation.