ACE-Step 1.5 in ComfyUI: setup and music generation

This tutorial covers ACE-Step 1.5 music generation in ComfyUI, using prompts and structured lyrics to create songs on the presenter's computer. The setup starts with updating ComfyUI and installing its Python requirements. The presenter encounters a startup error after skipping new dependencies and explains why that installation matters.

Two templates provide different loading options. The all-in-one workflow uses a single 10 GB checkpoint; the modular workflow loads text encoders, a diffusion model and the ACE 1.5 VAE separately. The demonstration uses the Turbo all-in-one model, sets the duration to 120 seconds and starts with eight sampling steps. Prompts describe instruments and vocal style, while bracketed lyric sections define verses and choruses. BPM and musical key have separate controls.

The presenter reports about four seconds of sampling for a two-minute track, then notes that text encoding and VAE decoding take additional time. The stated roughly 4 GB VRAM requirement is a speaker claim rather than a measured hardware comparison.

Tests include relaxed swing, Indian hip-hop fusion and K-pop. The presenter finds that some outputs follow the prompt well, but attempts to reproduce an official demo produce less clear lyrics. Changing seeds and increasing steps to 12 does not fully resolve that difference. The closing section points to LoRA training code in the repository without demonstrating training.