This LORA model is trained on an image of a sanpaku eyes (small iris).
V1.0
The iris becomes smaller when the weight is increased. However, if the weight is raised too high, the image becomes rough. 1.2 or lower is better.
The image of a boy has a smaller iris than that of a girl.
Angry, surprised, wide-eyed prompts work better.
For realistic checkpoints, an older, more realistic person is generated.
V2.0
In version 2, the training step was increased to generate more smaller iris while using the same input images.
LoRA weights can now be used with a weight of 1.
When adapted to realistic models, it produces younger person than version 1(why?).
Description
Increased the number of training repitisions from 2 to 5 and increased the number of steps from 1500 to 3750
FAQ
Comments (1)
is there a setting to make white space above the irises?
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.