|
--- |
|
license: creativeml-openrail-m |
|
tags: |
|
- stable-diffusion |
|
--- |
|
<center> <summary style="font-size: 2.25em; font-family: font-family: Segoe UI"><strong>🐰🌟 Bunnyspoon Alpha v1.0! 🌟🐰</h2></strong></summary></center> |
|
|
|
# <summary style="font-size: 2.25em; font-family: font-family: Segoe UI"><strong> |
|
Hi everyone! I've been an AI enthusiast since I know myself but started really going ham with the **Dall E 2** release. I've beta tested it and basically adopted Stable Diffusion day 1 when it was released. I have learned alot from the community and I want to share a bit back! Regards to the folks at **/hdg/**. |
|
</strong></summary> |
|
|
|
# <summary style="font-size: 2.25em; font-family: font-family: Segoe UI"><strong>♥ Donate</strong></summary> |
|
|
|
|
|
If you feel like it, I am accepting donations on |
|
|
|
- [Ko-Fi](https://ko-fi.com/lyloguum) |
|
- [Patreon (Nsfw)](patreon.com/Lylo011) |
|
|
|
<summary style="font-size: 2.25em; font-family: font-family: Segoe UI"><strong>⚙️ Recommended settings</strong></summary> |
|
|
|
<hr> |
|
<h1> |
|
|
|
<summary style="font-size: 2.25em; font-family: font-family: Segoe UI"><strong>🐰 Samples</strong></summary> |
|
|
|
## <h1 style="font-size: 1.5em; text-align: center; font-family: Segoe UI"> These are the settings I always use it is recommended but not required; |
|
| Settings | Value | |
|
| ----------------- | ------------------------------------------------------------------ | |
|
| Steps | 20+ | |
|
| Sampler | DPM++ 2M Karras | |
|
| CFG scale | 6 | |
|
| Size |576x768 | |
|
| Clip skip | 2 | |
|
| ENSD | 31337 | |
|
| Hires Fix | Enabled | |
|
| Upscale by | 1.5-1.9 | |
|
| Upscaler Fix | AnimeSharp | |
|
| DDetailer | Enabled | |
|
|
|
(install it as an extension on Automatic1111 WebUI) : https://github.com/DominikDoom/a1111-sd-webui-tagcomplete |
|
<hr> |
|
</h1> |
|
</h1> |
|
|
|
## <h1 style="font-size: 1.5em; text-align: center; font-family: Segoe UI"> Samples; </h1> |
|
<a href="https://i.imgur.com/rrzixBb.png"><img src="https://i.imgur.com/rrzixBb.png" width="550"></a> |
|
```c# |
|
Orihime Inoue, \(bleach series\), 1girl, mature female, adult, (old:1.3), medium Breasts, Athletic, abs, short hair, cone hair bun, |
|
split-color hair, aqua hair, green hair, aquarium, [Demon, monster, demon horns, red skin, 1boy, mature male, adult, old, Scrawny, thin:0.7], |
|
NSFW, 1girl, solo focus, Masterpiece, best quality, (style of star wars:1.2), pov, looking at viewer, loli, aged_down, Medium breasts, |
|
Skinny, Disgust, Narrowed eyes, begging, Transparent Sweater, Lederhosen, 20%frills, absurdly long hair, high ponytail, colored hair tips, light green hair, blonde hair |
|
``` |
|
|
|
```c# |
|
Negative |
|
Negative prompt: monochrome,(worst quality, low quality:1.4), zombie, EasyNegative,(multiple views:1.9),deleted, lowres, comic, bad anatomy,text, error,ugly, cropped, jpeg artifacts, signature, watermark, username, blurry,render,censored, |
|
bar censor, (multiple girls:1.9),render, 3d,realistic,(simple background:1.2), badhandv4,(upside-down:1.9),bokeh, covered nipples, nipples |
|
``` |
|
<a href="https://i.imgur.com/kk9WPR6.png"><img src="https://i.imgur.com/kk9WPR6.png" width="550"></a> |
|
<a href="https://i.imgur.com/WB1aWP9.png"><img src="https://i.imgur.com/WB1aWP9.png" width="550"></a> |
|
<a href="https://i.imgur.com/LB8wOQ5.png"><img src="https://i.imgur.com/LB8wOQ5.png" width="550"></a> |
|
|
|
<h1> |
|
|
|
<summary style="font-size: 2.25em; font-family: font-family: Segoe UI"><strong>📘 Cookbook</strong></summary> |
|
Bear with me. Most likely I will create a figma board to showcase this, if anything does not make sense feel free to ask. |
|
First of all, I really dig the kawacc mix model and I have seen it in action, I believe the model has a lot of potential (close to Counterfeit). So the goal here was to take the style of kawaccmix and improve on it. |
|
So, first of all I have done a style transfer via block merge from kawaccmix (model A) to csrb-yohan (model B): |
|
|
|
![](https://cdn-images.imagevenue.com/8a/58/1f/ME160CFE_o.png) (like this but reversed) |
|
----------- |
|
|
|
|
|
|
|
![kawaccmix](https://i.imgur.com/fZnKG9F.jpg) |
|
kawaccmix |
|
++ |
|
----------- |
|
|
|
![enter image description here](https://i.imgur.com/ypnXnPq.jpg) |
|
csrb yohan |
|
----------- |
|
|
|
|
|
This resulted in the following model, nicknamed csrb - yohaccmix: |
|
![enter image description here](https://i.imgur.com/663nKuq.jpg) |
|
----------- |
|
|
|
It successfully retained kawa's style and added it to yohan's anatomy which is better. |
|
|
|
Afterwards I have merged this with Anything v5 Ink with the purpose of increasing overall image quality (blocks below). |
|
![enter image description here](https://cdn-images.imagevenue.com/e8/35/3a/ME160CFG_o.png) |
|
|
|
![enter image description here](https://i.imgur.com/L7gCX4n.jpg) |
|
Nicknamed anyYohaCCMix v5. Using same blocks it was merged with [KuroGishikiv1.1](https://huggingface.co/Aotsuyu/Gishiki/blob/main/KuroGishikiv1.1.safetensors) CLIP/TXT Encoder retained |
|
----------- |
|
Result: |
|
![enter image description here](https://i.imgur.com/CmuRfMu.jpg) |
|
Nicknamed lylo megamix alpha v0.1. |
|
----------- |
|
Bear with me, next I have merged it with none other than Auora One (Huge thanks to Luna for his/her detailed instructions on figma, this guide and mix is a result of that). Seriously, [check out Aurora One](https://huggingface.co/SweetLuna/Aurora), in my personal opinion it is the best model overall and it has a huge versatility. |
|
|
|
I have used the same blocks as above, but have decided to go with Aurora's UNET/Text encoder with a 0-1 ratio. |
|
|
|
![enter image description here](https://i.imgur.com/IgSTj5s.jpg) |
|
Dubbed lylo-aurora |
|
----------- |
|
|
|
So from here on, you can pretty much mix this versatile model with anything you can think of. Personally I went with a modified personal version of [Mix-Pro-V4](https://civitai.com/models/7241/mix-pro-v4) (If there's demand for it I will share the model). |
|
|
|
Using The style blocks only but reversed (IN06-IN11 at 1, everything else 0), I have transferred the mix-pro-V4's modified style to lylo-aurora's composition. As you can see the results are stunning: |
|
|
|
UNET/TXTEncoder of 1 (lylo-aurora <--> mix-pro) |
|
![enter image description here](https://i.imgur.com/nOyQH2V.jpg) |
|
___________________ |
|
UNET/TXTEncoder of 0 (lylo-aurora <--> mix-pro) |
|
![enter image description here](https://i.imgur.com/rLJ9v5d.jpg) |
|
_________________________ |
|
|
|
|
|
You can delete the current file by clicking the **Remove** button in the file explorer. The file will be moved into the **Trash** folder and automatically deleted after 7 days of inactivity. |
|
|
|
**Lastly, the model was merged with [YuhMix A1](https://huggingface.co/lll-yuh-lll/YuhMix)**. |
|
| IN04-07 | 0.5 | |
|
|--|--| |
|
| Everything else | 0 | |
|
|
|
Finding those blocks was basically trial and error. IN04-07 affect main composition afaik and 0.5 was the sweet spot. |
|
![enter image description here](https://i.imgur.com/IgSTj5s.jpg) |
|
|
|
I will release the base model soon, just gotta nail the perfect settings first. However, when making Bunnyspoon I've decided to integrate 3 loras into it as well: |
|
|
|
- lora:alexi:.4 |
|
- lora:shirow-mix2-512d128a-768p-e08:.4 |
|
lora:hyocorou-wd-e04:.4 |
|
|
|
**FINAL RESULT** |
|
![enter image description here](https://i.imgur.com/zJipqNy.png) |
|
|
|
Hope you enjoyed! If you have any questions feel free to ask! |
|
|
|
|
|
<summary style="font-size: 2.25em; font-family: font-family: Segoe UI"><strong>⚡DISCLAIMER⚡</strong></summary> |
|
By downloading and using this model or any other model/mix based on this you assume full legal responsibility. The author is not and will not be responsible for any misguided use. |
|
|
|
|
|
# Love you ~! -lylo |