Sign In

[ssssloli] JK white pantyhose legs beautiful legs

34
436
96
12
Updated: Jun 21, 2024
clothingsexywomangirlsrealisticlegs
Verified:
SafeTensor
Type
LoRA
Stats
436
96
4
Reviews
Published
Jun 17, 2024
Base Model
SD 1.5
Trigger Words
1girl,
white pantyhose,
Pure girl style,
Slender legs,
Fat-free legs,
Hash
AutoV2
810E40614B
default creator card background decoration
ssssloli's Avatar
ssssloli

V1 version description:

This lora is mainly trained for the dressing style of lolita and Shaololita and the style of mirror selfie. It can perfectly optimize the legs and is more in line with the aesthetics of those who like thin legs and thin legs;

Because there are more mirror selfie effects, there will be problems when generating full-body pictures and characters whose faces need to be clear. It is not compatible with some characters' face lora for the time being. The subsequent V2-V3 version update will solve this problem.

Among them, the leg proportion and thin legs have been greatly optimized and processed, and basically there will be no meaty legs and thick legs similar to the European and American sense.

The calf proportion has been appropriately lengthened, and there is no fat and fat overall, so it looks thinner.

If you don't like thin legs, you can not use this lora, or reduce the weight of this lora.

————————————————————————————

In terms of clothing, it cooperates with the training of a variety of different lolita girl dressing styles, including maid clothes and Lolita styles. It can be understood as a powerful integrated version of lora.

Of course, there is no special training for shoes, but after testing, it is found that Mary Jane leather shoes can still be worn, and there are some furry slippers, which are very suitable for the current lolita style.

——————————————————————————

Problems with the V1 version for the time being:

1. Because the original training pictures come from the Internet, and most of them are the effects of selfies in front of the mirror without showing the face, even if there is no selfie or mobile phone in the keywords, there is still a probability that the effect of holding a mobile phone to block the face for selfies will appear, especially when the weight is higher than 0.7. The probability of occurrence is greater. The subsequent update version will solve this problem.

2. If you really don’t want to block the face with a mobile phone in hand, it is recommended to add keywords such as selfie, holding a mobile phone in hand, and mobile phone in the negative words and give these words some weight; then appropriately reduce the weight of lora, and there is a high probability that the mobile phone will not appear.

3. It is not compatible with most face lora for the time being. Of course, you can generate a picture without the phone blocking the face. Although the face may be distorted and strange at this time, you can turn off this lora in the partial redrawing of the picture, load your character face lora, and generate a new face to replace it. The actual test effect is very nice.

4. Because a large number of trainings are standing postures, when you use keywords of other postures, there may be structural confusion problems. The V3 version will be gradually updated to allow the lora to adapt to more postures and postures.

If you have other questions, please add them in the comment area. The V1 version is indeed not perfect, but everyone looks forward to it and it will do better.

Of course, the current leg optimization and the loli legs in white silk stockings that you want are already there, right?

————————————————————————————

Details of V1 version parameters:

Large model recommendation: ChilloutMix_Chilloutmix-Ni-pruned-fp32-fix

Sampling method: DPM++2M Karras (you can also try others)

steps: 20-40

High-resolution repair: recommended to enable; Enlargement algorithm: R-EXRGAN 4x+; redraw amplitude: 0.2-0.3;

CFG: 3-7

Lora weight recommendation: 0.5-0.9

————————————————————————————

Lora weight details:

Lora weight 0.6 with corresponding trigger words: leg proportions and details are optimized to a certain extent, clothing style is not high, and other content generation effects are more random;

Lora weight 0.8 with corresponding trigger words: leg proportions and details are greatly optimized, clothing style is closer to Lora training, but it is easy to block the face with a mobile phone in hand;

So everyone can adjust the weight of Lora within an appropriate range according to their needs.