25 photos were learned 40 times each for a total of 10 rounds. The underlying model for training was not used Pony DiffusionV6XL_v6StartWithThisOne , and the model chosen for use was a variety of other randomly selected large models.

@tova20200101639 ah and i also another question,
how did you get these images as a dataset? that steel panties doesn't exist for real

@sienwn Using the model of sdxl with some metallic lora, after reaching a satisfactory effect by trying the effect a little bit, the most accurate expression is selected from a large number of data, which can be based on the number of dozens to thousands of photos, which are all possible. When these concepts are determined, the previous content through the complex description and complex lora combinations are made into an instrumental Lora, as long as the use of this one Lora, the concept of the previous complex description can be achieved with only a few keywords.

@tova20200101639 thanks, you used a image from SDXL , i thought you draw that panties by hands.

@sienwn If I had drawing skills, I'd prefer to draw some of the content myself and then implement the various concepts lora.