注意:Z-Image Turbo使用多個LORA時圖片容易出現燒焦錯誤的情況。
Note: When using multiple LoRa images with Z-Image Turbo, images are prone to burning errors.
--
2025/12/05:
A lora或B lora可單獨使用，A訓練步驟較低，B訓練步驟較高，當A和B一起使用時只需LORA權重0.4就可生效。
但是要組合和其它LORA時，則必須A或B單獨使用。
此LORA會影響人臉，如果要還原內建人物的臉則必須進行面部修復。
或是使用A和B一起使用並且權重0.4即可降低對內建人物臉部的影響。
使用H200 SXM訓練。
--
Lora A or Lora B can be used individually. Lora A has a lower training step count, while Lora B has a higher training step count. When A and B are used together, a Lora weight of 0.4 is sufficient for effectiveness.
However, to combine them with other Lora types, either A or B must be used individually.
This Lora type affects the face; facial retouching is necessary to recreate the built-in character's face.
Alternatively, using A and B together with a weight of 0.4 can reduce the impact on the built-in character's face.
Training is performed using H200 SXM.
--