“vore”“same_size”“No endoscopy”
The trigger phrase is "distended belly outline in same size vore, swollen abdomen contour, illustration style." This is the only example; please see the image I uploaded.
Suitable for "Gender: Female," "Race: Human."
Description
FAQ
Comments (4)
testing v2 vs v1, v2 is much stronger than v1 by about 3 times. You get similar results from v2 at .35 strength as you would with v1 at strength 1.
It seems it was only trained on anime cartoon images as the both push the image in that direction. That being good or bad is a matter of preference, but just making clear what it is.
进阶
优化方向，减少lora的污染性
使用“Selective LoRA Loader (Z-Image)”节点进行分层
”stregtn“权重0.8附近
分为以下两种方案
0-9修改为0.3其他1
0-9修改为0.3,10-14修改为0.6其他1
Advanced
Optimization direction: Reduce LoRA's contamination
Use the "Selective LoRA Loader (Z-Image)" node for layering
"strength" weight around 0.8
Divided into the following two schemes:
Modify 0-9 to 0.3, others to 1
Modify 0-9 to 0.3, 10-14 to 0.6, others to 1
Is ZIT better than SDXL-based models at generating content such as vore?
I opened two privately, and the performance of fur as a dataset can only be described as very poor. The weights in the model are too low, and I don't think this is a problem that Lora can solve. At least we need to make a slight adjustment to the zimage first, and then train it to achieve good results