You can use Strength 1.0 only when using it on its own. However, Z‑image Turbo currently causes Lora models to interfere with each other quite heavily, so set it to 0.5 when stacking multiple Loras. Nobody will read this anyway so it will probably be ignored, but things may change once training Loras directly on the base model becomes possible. For now, the countermeasures are limited, so when stacking Loras, keep the Strength at a maximum of 0.6.

To explain in more detail: Z‑image is a lightweight model with a small number of layers, which means its layers overlap heavily with those of other Loras. Because the main content is concentrated in specific layers, interference between Loras occurs very easily. There are several countermeasures, but under the current conditions they only provide limited effectiveness. Unlike large models such as Flux or Qwen, Z‑image does not have much tolerance, so if you stack too many Loras on top of each other, the entire output will inevitably collapse.