ACE-Step Workflow: Negative Prompts Are Key to Predictable Background Music

Illustrious_Usual_10 · reddit · 2026-08-11

A developer shared a ComfyUI and ACE-Step workflow for generating deliberately predictable background music, aimed at listeners (like those with ADHD or autism) who require high focus.

The key insight is that the negative prompt does all the heavy lifting. To prevent the model from naturally trying to add structure, you must explicitly ban terms like 'crescendo', 'build up', 'drop', and 'dynamic change'.

Other practical tips include keeping generations under 150 seconds to avoid the model losing the thread, locking the seed before experimenting, and using a shell script to fold the track for seamless looping.

Original post →

More from Multimodal

Multimodal channel →