LoRA strength in ComfyUI scales the weight delta additively and can be negative or greater than 1

The LoraLoader node applies a LoRA to MODEL and CLIP separately with independent strength_model and strength_clip sliders (default 1.0). Internally, strength multiplies the calculated weight delta (lora_B @ lora_A * alpha/rank) before adding it to the base weight. Values between 0.5–0.9 blend the LoRA subtly; 1.0 applies full strength; >1.0 exaggerates the style. Negative values subtract the LoRA’s influence (style suppression). Setting strength_model=1.0 and strength_clip=0.0 applies the visual style without changing the text encoder.

Examples

strength_model=0.7, strength_clip=0.7 → moderate style blend; strength_model=1.2 → exaggerated style; strength_model=-0.5 → suppress the style.

Assessment

A user wants to use a character LoRA for its face details but their prompts are already well-tuned to the base model. Which strength combination should they try and why?