Ace-Step-1.5-XL-Concept-Sliders

A collection of experimental ACE Step 1.5 XL concept slider LoRAs for pushing generations toward or away from specific musical traits, vocal styles, production styles, and arrangement characteristics.

These are directional sliders, so the effect depends on the LoRA strength:

For files with two concepts in the name, such as male-female, the first word is the positive direction and the second word is the negative direction. Example: male-female means positive strength pushes toward male vocals, while negative strength pushes toward female vocals.

For files with one concept in the name, such as happiness, positive strength adds more of that concept and negative strength reduces it.

Recommended usage: Start with 1.0 or -1.0, then adjust from there. Some sliders may need stronger values like 2.0 or higher, while others may work better around 0.5 or even 0.2. In general, 1.0 is a good starting point for testing the direction.

Training note: These were trained using the concept slider training option in AI Toolkit. They are not traditional concept LoRAs trained on curated examples of each specific trait. The slider method uses prompt directions, such as one concept as the positive direction and another as the negative direction, so it does not require a dedicated dataset for every individual slider concept. This made it possible to create a wide range of small directional LoRAs from the same general audio training setup.