Ace-Step-1.5-XL-Concept-Sliders Dial Up Fine-Grained AI Music Control

The Ace-Step-1.5-XL-Concept-Sliders are a new set of directional LoRA files that let you nudge an AI music generator toward or away from specific audio traits. These sliders work with the Ace-Step-1.5-XL model, adjusting elements like vocal type, production style, and mood with a simple strength setting. Dial a positive value to add a trait, or a negative one to steer in the opposite direction.

Independent developer Xanthius created the sliders by modifying the AI Toolkit software to support this type of training. He built a wide collection without needing separate example datasets for each trait, relying instead on prompt direction pairs. The result is a grab bag of adjustable dials that give users much finer control over music generations.

A toolkit of sound-shaping controls

Music producers using local AI tools can quickly iterate on a track’s feel without rewriting lengthy prompts. Hobbyist musicians who want to experiment with vocal styles or production vibes will find these sliders an easy starting point. Because everything runs locally, privacy-conscious users can shape audio without sending ideas to the cloud.

How the sliders were built

Xanthius notes that AI Toolkit lacked native support for Ace-Step slider LoRAs, so he edited the code to make training possible. Each slider took between ten minutes and an hour to create, using prompt opposites instead of curated example datasets. He plans to expand the collection based on community suggestions and is open to new trait ideas.

"I was able to edit the code enough to get it working properly and now I can train concept sliders in about 10 mins to an hour each and without needing specific datasets for the concepts." — Source: Reddit