ComfyUI Workflow: AI Music Generation with Ace-Step 1.5

This ComfyUI workflow generates complete music tracks from a text prompt using Ace-Step 1.5, running locally from a single checkpoint file. Describe the genre, mood, and instrumentation you want, and the model produces a full track: drums, bass, chords, and melody, mixed and ready to export as audio.

How It Works

The graph is a standard diffusion setup tuned for audio:

• TextEncodeAceStepAudio1.5 turns your prompt into conditioning. Describe the genre, tempo, instrumentation, and vocal style in plain language, the same way you would prompt an image model.
• EmptyAceStep1.5LatentAudio provides the empty audio latent the sampler refines into sound.
• KSampler runs the diffusion steps, with a model-sampling node setting the correct schedule for the audio model.
• VAEDecodeAudio and SaveAudioAdvanced decode and write the finished track to file.

Quick Start in ComfyUI

• Load the workflow using the download button on this page.
• Download the checkpoint: ace_step_1.5_turbo_aio.safetensors from the official Comfy-Org Hugging Face repository, and place it in models/checkpoints/. This one file contains everything the workflow needs, which is why the turbo all-in-one version is the easiest starting point.
• Write a prompt that names the genre and the feel. The template ships with a neo-soul example prompt you can rewrite.
• Run the queue and listen. Longer generation times produce longer tracks.

Good To Know

• The turbo checkpoint trades a little quality for speed. Once the base graph works, the XL variant with split model files is the upgrade path.
• Update ComfyUI to the latest version first; the Ace-Step nodes are recent additions and older builds do not include them.
• Prompt with musical vocabulary: "warm neo-soul with live drums", "fast drum and bass, no vocals". Genre words steer the output more than adjectives do.

Why Use This Workflow

• Generate royalty-free beds for videos, games, and podcasts without digging through stock libraries.
• Everything runs locally from one model file, with no subscription, no API key, and no per-track cost.
• Because it sits in ComfyUI, you can chain it: generate a track, then drive a video workflow from the same session.