BakaOnE / README.md
kgmkm's picture
Update README.md
625e952
---
license: creativeml-openrail-m
pipeline_tag: text-to-image
---
![top](https://huggingface.co/kgmkm/BakaOnE/resolve/main/image/1-1.jpg)
There are three models(*1) that I like, and I have carefully merged them into "BakaOnE".
![top](https://huggingface.co/kgmkm/BakaOnE/resolve/main/image/2-1.jpg)
By taking advantage of these original trained model(*2), we have achieved a quality that is as good as other sd1.5 models.
![top](https://huggingface.co/kgmkm/BakaOnE/resolve/main/image/2-2.jpg)
When hierarchically merging this model, we used multiple options to select limbs and hands that were stable. Additionally, I followed the inspiration from my testicles and chose the better result.
Did you notice how many latex bodysuits and swimsuits are in the sample images? That's what I mean.
![top](https://huggingface.co/kgmkm/BakaOnE/resolve/main/image/2-4.jpg)
*1: sorry, my goal is "DONT USED NovelAI LEAK", but i wrong.
*2: those licence is CreativeML Open RAIL-M. it is too.
## recommend vae
this model isnt baked. use vae others.
my highly recommend vae is vae-ft-mse-840000-ema-pruned.safetensors :
https://huggingface.co/stabilityai/sd-vae-ft-mse-original/tree/main
## notice: sample pics
i highly recommend the model used with lora.
i like lora for visual effect, especially made 2vXpSwA7's lora (ex: <lora:flat2:-0.5>, <lora:hairdetailer:0.4>, <lora:deluster1:-0.5>)
https://huggingface.co/2vXpSwA7/iroiro-lora
The second half of the pics shows examples of these uses.
and this:
dont used
![top](https://huggingface.co/kgmkm/BakaOnE/resolve/main/image/3-1.jpg)
used
![top](https://huggingface.co/kgmkm/BakaOnE/resolve/main/image/3-2.jpg)