Update README.md
Browse files
README.md
CHANGED
@@ -7,32 +7,67 @@ tags:
|
|
7 |
language:
|
8 |
- en
|
9 |
---
|
10 |
-
<center><h1><b>
|
11 |
-
<center><img src="https://huggingface.co/Aotsuyu/Kukicha/resolve/main/images/
|
12 |
|
|
|
|
|
|
|
|
|
13 |
|
14 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
|
16 |
-
HoloKukicha is just the same model but with Vtubers merged in but it's what I mostly use so grab that.
|
17 |
|
18 |
### Defaults that I use:
|
19 |
|
20 |
```
|
21 |
-
detailed background, masterpiece, best quality,
|
22 |
Negative prompt: (low quality, worst quality:1.4), (bad anatomy), extra digit, fewer digits, (extra arms:1.2), bad hands, by (bad-artist:0.6), bad-image-v2-39000
|
23 |
-
Steps: 25, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1189140555, Size: 512x768,
|
24 |
```
|
25 |
This is with kl-f8-anime2 vae, probably use higher cfg if trying with the anything or whatever vae, hell, mess around.
|
26 |
|
27 |
-
I also sometimes use 22ish cfg with:
|
28 |
```
|
29 |
Dynamic thresholding enable#: True, Mimic scale: 7, Threshold percentile: 96, Mimic mode: Half Cosine Up, Mimic scale minimum: 4, CFG mode: Half Cosine Up, CFG scale minimum: 4
|
30 |
```
|
31 |
|
|
|
32 |
### DARKNESS
|
33 |
The darkKukiv1.safetensors LoRA included in the repo is my first version of a noise offset LoRA specifically for this model.
|
34 |
|
35 |
-
I'm working on improved
|
36 |
What it does is it makes the model able to output a wider range of lighting, example:
|
37 |
|
38 |
<center><img src="https://i.imgur.com/ixZWK6N.png"width="55%"/></center>
|
@@ -47,13 +82,17 @@ This is an another source of noise offset so if that's an issue, disable the dar
|
|
47 |
|
48 |
And just read the readme of the LoRA.
|
49 |
|
50 |
-
|
51 |
https://huggingface.co/Aotsuyu/HorrorLora/
|
52 |
|
53 |
|
54 |
-
|
|
|
|
|
|
|
55 |
|
|
|
56 |
|
|
|
57 |
|
58 |
- [Closertodeath](https://huggingface.co/closertodeath) for being cute.
|
59 |
- [andite](https://huggingface.co/andite) for being handsome.
|
@@ -61,10 +100,3 @@ https://huggingface.co/Aotsuyu/HorrorLora/
|
|
61 |
- [NoCrypt](https://huggingface.co/NoCrypt) for providing the necessary computing power.
|
62 |
- Other members of the 東方Project AI community.
|
63 |
- Anons responsbile for hololive/vtuber finetunes.
|
64 |
-
|
65 |
-
#### This feels shitpost-y
|
66 |
-
That's because it is, writing readmes is hard.
|
67 |
-
|
68 |
-
I might be arsed to improve this model card, upload fp16 and put this all on civit at some point.
|
69 |
-
|
70 |
-
That point isn't now though.
|
|
|
7 |
language:
|
8 |
- en
|
9 |
---
|
10 |
+
<center><h1><b>HoloKuki</b></h1></center>
|
11 |
+
<center><img src="https://huggingface.co/Aotsuyu/Kukicha/resolve/main/images/nakiricool.png"width="25%"/></center>
|
12 |
|
13 |
+
### Previews!
|
14 |
+
To make the previews fair I applied no editing, no inpainting or such techniques.
|
15 |
+
All the images have metadata.
|
16 |
+
There might be more previews on the [civitai](https://civitai.com/models/17598/holokuki) page, since it's easier to upload them there.
|
17 |
|
18 |
+
<center><img src="https://huggingface.co/Aotsuyu/Kukicha/resolve/main/images/widekiara.png"></center>
|
19 |
+
<center><img src="https://huggingface.co/Aotsuyu/Kukicha/resolve/main/images/suisei.png"></center>
|
20 |
+
<center><img src="https://huggingface.co/Aotsuyu/Kukicha/resolve/main/images/reimu.png"></center>
|
21 |
+
<center><img src="https://huggingface.co/Aotsuyu/Kukicha/resolve/main/images/nakiriwide.png"></center>
|
22 |
+
<center><img src="https://huggingface.co/Aotsuyu/Kukicha/resolve/main/images/nakiri.png"></center>
|
23 |
+
<center><img src="https://huggingface.co/Aotsuyu/Kukicha/resolve/main/images/miku.png"></center>
|
24 |
+
<center><img src="https://huggingface.co/Aotsuyu/Kukicha/resolve/main/images/kiara!.png"></center>
|
25 |
+
<center><img src="https://huggingface.co/Aotsuyu/Kukicha/resolve/main/images/sceneryday.png"></center>
|
26 |
+
<center><img src="https://huggingface.co/Aotsuyu/Kukicha/resolve/main/images/scenerynight.png"></center>
|
27 |
+
<center><img src="https://huggingface.co/Aotsuyu/Kukicha/resolve/main/images/pekosunflower.png"></center>
|
28 |
+
|
29 |
+
### Holos!
|
30 |
+
This is a big ass file, open it and look at the holos. Stare at them.
|
31 |
+
<center><img src="https://huggingface.co/Aotsuyu/Kukicha/resolve/main/images/xyz_grid-0027-541571007-detailed%20background%2C%20masterpiece%2C%20best%20quality%2C%20(tokino%20sora_1.2)%2C%20blue%20eyes%2C%20brown%20hair%2C%20long%20hair%2C%20hair%20flaps%2C%20hairclip%2C%20red%20r.png"></center>
|
32 |
+
|
33 |
+
|
34 |
+
### Description
|
35 |
+
|
36 |
+
TL;DR WHAT I TRIED TO ACHIEVE:
|
37 |
+
A midpoint between being able to generate stunning scenery (dpep) and just cute anime girls (Anything) with some other models mixed in - to help improve other aspects.
|
38 |
+
|
39 |
+
It's a model, made by mixing other models, the base were Anythingv4.5 and dpep-chillout mix.
|
40 |
+
I've made many mixes before and I've applied the knowledge I've gotten from that here - a lot of BW merging was applied.
|
41 |
+
|
42 |
+
I feel like this mix is good enough to contend with the most popular ones, and it's the one I use currently, for all my purposes.
|
43 |
+
|
44 |
+
It can do SFW, NSFW, thanks to closer to death's dpep it has the capability for interesting backgrounds and I've merged in a VTuber model - all of Hololive can be prompted for.
|
45 |
+
But I think the images speak for itself better than I can.
|
46 |
+
|
47 |
+
|
48 |
+
### What to grab
|
49 |
+
I advise grabbing [this file](https://huggingface.co/Aotsuyu/Kukicha/blob/main/HoloKukiv2-fp16.safetensors) - this is version 2.
|
50 |
|
|
|
51 |
|
52 |
### Defaults that I use:
|
53 |
|
54 |
```
|
55 |
+
detailed background, masterpiece, best quality, prompty promptu
|
56 |
Negative prompt: (low quality, worst quality:1.4), (bad anatomy), extra digit, fewer digits, (extra arms:1.2), bad hands, by (bad-artist:0.6), bad-image-v2-39000
|
57 |
+
Steps: 25, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1189140555, Size: 512x768, Clip skip: 2, ENSD: 31337, Discard penultimate sigma: True
|
58 |
```
|
59 |
This is with kl-f8-anime2 vae, probably use higher cfg if trying with the anything or whatever vae, hell, mess around.
|
60 |
|
61 |
+
I also sometimes use 22ish cfg with the dynamic thresholding extension:
|
62 |
```
|
63 |
Dynamic thresholding enable#: True, Mimic scale: 7, Threshold percentile: 96, Mimic mode: Half Cosine Up, Mimic scale minimum: 4, CFG mode: Half Cosine Up, CFG scale minimum: 4
|
64 |
```
|
65 |
|
66 |
+
|
67 |
### DARKNESS
|
68 |
The darkKukiv1.safetensors LoRA included in the repo is my first version of a noise offset LoRA specifically for this model.
|
69 |
|
70 |
+
I'm working on improved version as well, but this one isn't bad by any means.
|
71 |
What it does is it makes the model able to output a wider range of lighting, example:
|
72 |
|
73 |
<center><img src="https://i.imgur.com/ixZWK6N.png"width="55%"/></center>
|
|
|
82 |
|
83 |
And just read the readme of the LoRA.
|
84 |
|
|
|
85 |
https://huggingface.co/Aotsuyu/HorrorLora/
|
86 |
|
87 |
|
88 |
+
### I want to know more about such and such
|
89 |
+
|
90 |
+
Your best bet is this https://rentry.org/TohoAIFAQ
|
91 |
+
I'm trying to slowly fill it out.
|
92 |
|
93 |
+
## Big Thanks to
|
94 |
|
95 |
+
You should also check out their models!
|
96 |
|
97 |
- [Closertodeath](https://huggingface.co/closertodeath) for being cute.
|
98 |
- [andite](https://huggingface.co/andite) for being handsome.
|
|
|
100 |
- [NoCrypt](https://huggingface.co/NoCrypt) for providing the necessary computing power.
|
101 |
- Other members of the 東方Project AI community.
|
102 |
- Anons responsbile for hololive/vtuber finetunes.
|
|
|
|
|
|
|
|
|
|
|
|
|
|