|
--- |
|
license: creativeml-openrail-m |
|
library_name: diffusers |
|
pipeline_tag: text-to-image |
|
tags: |
|
- ' stable-diffusion' |
|
- stable-diffusion-diffusers |
|
--- |
|
# what is "SSSSLLDDLL" |
|
check prompt for this pics. that's all. |
|
|
|
> masterpiece, shiny sissy luxury latex dress for doll, (kawaii:1.3), ultimate quality, |
|
|
|
> Negative prompt: (worst_quality:1.4), (inaccurate limbs:1.4), (twisted finger:1.4), poorly=drawn, (lowres:1.3), splatoon, tall height, animal ears, |
|
|
|
 |
|
|
|
|
|
|
|
日本語はこちら https://huggingface.co/kgmkm/SSSSLLDDLL/blob/main/README-ja.md |
|
|
|
|
|
|
|
# Overview |
|
|
|
I met a model who expressed beautiful latex costumes, and I wanted to create an anime-style model while maintaining this gloss. In my opinion, the traditional 2.5-dimensional style has many illustrations inspired expressions. Therefore, I adjusted the merge model to be realistic overall but anime-style in the face, referring to figures and dolls. |
|
|
|
There are only two main purposes for the merge. These two axes may be contradictory or synergistic. Carefully adjusting these while discovering new ideas, I merged them in the Hierarchical LoRA. |
|
|
|
## 1. Make it cute and ANIME-like |
|
|
|
- Especially make the eyes cuter than traditional models |
|
- Adopt cute decorative models if possible |
|
|
|
These two are emphasized. |
|
|
|
 |
|
|
|
By putting the magic spell "kawaii: 1.2" the face changes depending on whether it is a girlish or adult style. As written in the old Japanese book "古事記", "kawaii is justice." |
|
|
|
 |
|
|
|
 |
|
|
|
## 2. Express the beautiful latex clothing |
|
|
|
- Enhance the expression of light so that the gloss of latex becomes more beautiful |
|
- Enrich the texture of the human body so that tight clothing looks good (I think a slightly plump girl looks better in a latex dress than a skinny one) |
|
|
|
I take photos for both work and hobby. From my experience photographing women and figures, I believe that the expression of light and shadow is the most important aspect of glossy clothing. Therefore, I adjusted the model using prompts that frequently use light expressions in photography. As a result, expressions such as blur, backlighting, bounce photography, environmental light, and blur became richer. |
|
|
|
 |
|
|
|
The above photo was taken using CFG 8, with a focus on strong lighting. This model can also produce softer textures. Please use CFG 3. CFG can be used satisfactorily up to about 2.5. |
|
|
|
 |
|
|
|
|
|
|
|
# Unexpected Range of Expressions |
|
|
|
By keeping these two main purposes in mind, we aimed to create a model that best represents "shiny sissy luxury latex dress for doll." However, as a side effect, other expressions also became beautiful and have become our favorites. |
|
|
|
## Texture |
|
|
|
Despite pursuing the expression of latex material, other fabrics such as satin and spandex are now expressed with rich tones. |
|
|
|
Example of "satin". "See-through" effect is also well-utilized. |
|
|
|
 |
|
|
|
Example of "spandex". Wet feeling is also well-expressed, and even the water is surprisingly portrayed realistically. |
|
|
|
 |
|
|
|
What was interesting was the example of silk. This model seems to consider silk to be something that can only be calculated in China. |
|
|
|
 |
|
|
|
## Other things |
|
|
|
Furthermore, the merged models were often strong in fantasy expressions, and it seems that fantasy and SF expressions are stronger than realistic ones. And by benefiting from the expression of light, it seems that even more artistic works can be created. |
|
|
|
The way light hits, the texture of metal, and the beauty of feathers were all beyond our imagination. The expression, like dust, was greatly influenced by UniPC. The charm of UniPC lies in the detail and dustiness of patterns such as wallpapers and backgrounds. On the other hand, SDE++ has a clean depiction. |
|
|
|
 |
|
|
|
Portraits can now be beautifully expressed, and we were able to reproduce the common photographic technique of bokeh. If there is anything in these images that catches your attention, please check the prompts. |
|
|
|
 |
|
|
|
## VAE |
|
|
|
no include VAE. |
|
|
|
We recommend using clearvae. This is the VAE that produces the most detailed image representations. However, it often generates blacked-out images. Because of this, it is not suitable for fixing the seed when using i2i or controlnet. |
|
|
|
Using vae-ft-mse-840000-ema-pruned.safetensors as an alternative is also a good option. |
|
|
|
## GLHF |
|
|
|
We have created a model that can be enjoyed with simple prompts and complex prompts like the above. But first, so simply, enjoy the "shiny sissy luxury latex dress for doll"! |
|
|
|
> masterpiece, shiny sissy luxury latex dress for doll, ultimate quality, |
|
|
|
 |
|
|
|
|
|
|
|
--- |
|
license: creativeml-openrail-m |
|
--- |
