kayfahaarukku
commited on
Update README.md
Browse files
README.md
CHANGED
@@ -1,5 +1,243 @@
|
|
1 |
-
---
|
2 |
-
license: other
|
3 |
-
license_name: faipl
|
4 |
-
license_link: https://freedevproject.org/faipl-1.0-sd
|
5 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: other
|
3 |
+
license_name: faipl
|
4 |
+
license_link: https://freedevproject.org/faipl-1.0-sd
|
5 |
+
language:
|
6 |
+
- en
|
7 |
+
tags:
|
8 |
+
- text-to-image
|
9 |
+
- stable-diffusion
|
10 |
+
- safetensors
|
11 |
+
- stable-diffusion-xl
|
12 |
+
base_model: cagliostrolab/animagine-xl-3.1
|
13 |
+
widget:
|
14 |
+
- text: >-
|
15 |
+
1girl, green hair, sweater, looking at viewer, upper body, beanie,
|
16 |
+
outdoors, night, turtleneck, masterpiece, best quality
|
17 |
+
parameter:
|
18 |
+
negative_prompt: >-
|
19 |
+
nsfw, lowres, bad anatomy, bad hands, text, error, missing fingers,
|
20 |
+
extra digit, fewer digits, cropped, worst quality, low quality, normal
|
21 |
+
quality, jpeg artifacts, signature, watermark, username, blurry, artist
|
22 |
+
name
|
23 |
+
example_title: 1girl
|
24 |
+
---
|
25 |
+
<style>
|
26 |
+
.title-container {
|
27 |
+
display: flex;
|
28 |
+
justify-content: center;
|
29 |
+
align-items: center;
|
30 |
+
height: 100vh; /* Adjust this value to position the title vertically */
|
31 |
+
}
|
32 |
+
|
33 |
+
.title {
|
34 |
+
font-size: 2.5em;
|
35 |
+
text-align: center;
|
36 |
+
color: #333;
|
37 |
+
font-family: 'Helvetica Neue', sans-serif;
|
38 |
+
text-transform: uppercase;
|
39 |
+
letter-spacing: 0.1em;
|
40 |
+
padding: 0.5em 0;
|
41 |
+
background: transparent;
|
42 |
+
}
|
43 |
+
|
44 |
+
.title span {
|
45 |
+
background: -webkit-linear-gradient(45deg, #bdabe3, #b39a3e);
|
46 |
+
-webkit-background-clip: text;
|
47 |
+
-webkit-text-fill-color: transparent;
|
48 |
+
}
|
49 |
+
|
50 |
+
.custom-table {
|
51 |
+
table-layout: fixed;
|
52 |
+
width: 100%;
|
53 |
+
border-collapse: collapse;
|
54 |
+
margin-top: 2em;
|
55 |
+
}
|
56 |
+
|
57 |
+
.custom-table td {
|
58 |
+
width: 50%;
|
59 |
+
vertical-align: top;
|
60 |
+
padding: 10px;
|
61 |
+
box-shadow: 0px 0px 0px 0px rgba(0, 0, 0, 0.15);
|
62 |
+
}
|
63 |
+
.custom-image-container {
|
64 |
+
position: relative;
|
65 |
+
width: 100%;
|
66 |
+
margin-bottom: 0em;
|
67 |
+
overflow: hidden;
|
68 |
+
border-radius: 10px;
|
69 |
+
transition: transform .7s;
|
70 |
+
/* Smooth transition for the container */
|
71 |
+
}
|
72 |
+
.custom-image-container:hover {
|
73 |
+
transform: scale(1.05);
|
74 |
+
filter: none;
|
75 |
+
/* Scale the container on hover */
|
76 |
+
}
|
77 |
+
.custom-image {
|
78 |
+
width: 100%;
|
79 |
+
height: auto;
|
80 |
+
object-fit: cover;
|
81 |
+
border-radius: 10px;
|
82 |
+
transition: transform .7s;
|
83 |
+
margin-bottom: 0em;
|
84 |
+
}
|
85 |
+
.nsfw-filter {
|
86 |
+
filter: blur(8px); /* Apply a blur effect */
|
87 |
+
transition: filter 0.3s ease; /* Smooth transition for the blur effect */
|
88 |
+
}
|
89 |
+
|
90 |
+
.overlay {
|
91 |
+
position: absolute;
|
92 |
+
bottom: 0;
|
93 |
+
left: 0;
|
94 |
+
right: 0;
|
95 |
+
color: white;
|
96 |
+
width: 100%;
|
97 |
+
height: 40%;
|
98 |
+
display: flex;
|
99 |
+
flex-direction: column;
|
100 |
+
justify-content: center;
|
101 |
+
align-items: center;
|
102 |
+
font-size: 1vw;
|
103 |
+
font-style: bold;
|
104 |
+
text-align: center;
|
105 |
+
opacity: 0;
|
106 |
+
/* Keep the text fully opaque */
|
107 |
+
background: linear-gradient(0deg, rgba(0, 0, 0, 0.8) 60%, rgba(0, 0, 0, 0) 100%);
|
108 |
+
transition: opacity .5s;
|
109 |
+
}
|
110 |
+
.custom-image-container:hover .overlay {
|
111 |
+
opacity: 1;
|
112 |
+
}
|
113 |
+
.overlay-text {
|
114 |
+
background: linear-gradient(45deg, #7ed56f, #28b485);
|
115 |
+
-webkit-background-clip: text;
|
116 |
+
color: transparent;
|
117 |
+
text-shadow: 2px 2px 4px rgba(0, 0, 0, 0.7);
|
118 |
+
|
119 |
+
.overlay-subtext {
|
120 |
+
font-size: 0.75em;
|
121 |
+
margin-top: 0.5em;
|
122 |
+
font-style: italic;
|
123 |
+
}
|
124 |
+
|
125 |
+
.overlay,
|
126 |
+
.overlay-subtext {
|
127 |
+
text-shadow: 2px 2px 4px rgba(0, 0, 0, 0.5);
|
128 |
+
}
|
129 |
+
|
130 |
+
</style>
|
131 |
+
<h1 class="title">
|
132 |
+
<span>UrangDiffusion 1.1</span>
|
133 |
+
</h1>
|
134 |
+
<table class="custom-table">
|
135 |
+
<tr>
|
136 |
+
<td>
|
137 |
+
<div class="custom-image-container">
|
138 |
+
<img class="custom-image" src="https://cdn-uploads.huggingface.co/production/uploads/64333a074521083b9d2aab3b/9OsyB_K1qMV99MPPMswzL.png" alt="sample1">
|
139 |
+
</div>
|
140 |
+
<div class="custom-image-container">
|
141 |
+
<img class="custom-image" src="https://cdn-uploads.huggingface.co/production/uploads/64333a074521083b9d2aab3b/tlBoncBJJg4ob7IEAip3j.png" alt="sample4">
|
142 |
+
</div>
|
143 |
+
</td>
|
144 |
+
<td>
|
145 |
+
<div class="custom-image-container">
|
146 |
+
<img class="custom-image" src="https://cdn-uploads.huggingface.co/production/uploads/64333a074521083b9d2aab3b/wAcN5GrWDJkCbYvBaFLKb.png" alt="sample2">
|
147 |
+
</div>
|
148 |
+
<div class="custom-image-container">
|
149 |
+
<img class="custom-image" src="https://cdn-uploads.huggingface.co/production/uploads/64333a074521083b9d2aab3b/3TrYflrSEZKqrCxarCN8H.png" alt="sample3">
|
150 |
+
</td>
|
151 |
+
<td>
|
152 |
+
<div class="custom-image-container">
|
153 |
+
<img class="custom-image" src="https://cdn-uploads.huggingface.co/production/uploads/64333a074521083b9d2aab3b/QKplYPDqVh--cIto3KdWS.png" alt="sample1">
|
154 |
+
</div>
|
155 |
+
<div class="custom-image-container">
|
156 |
+
<img class="custom-image" src="https://cdn-uploads.huggingface.co/production/uploads/64333a074521083b9d2aab3b/xRZB2_FDDWmPj_jVj_PHy.png" alt="sample4">
|
157 |
+
</div>
|
158 |
+
</td>
|
159 |
+
</tr>
|
160 |
+
</table>
|
161 |
+
|
162 |
+
**UrangDiffusion 1.2** (oo-raw-ng Diffusion) is an updated version of UrangDiffusion 1.1. This version provides dataset refresh, improvements over the last iteration and training parameter correction.
|
163 |
+
|
164 |
+
## Standard Prompting Guidelines
|
165 |
+
The model is finetuned from Animagine XL 3.1. However, there is a little bit changes on dataset captioning, therefore there is some different default prompt used:
|
166 |
+
|
167 |
+
**Default prompt**:
|
168 |
+
```
|
169 |
+
1girl/1boy, character name, from what series, everything else in any order, masterpiece, best quality, amazing quality, very aesthetic
|
170 |
+
```
|
171 |
+
|
172 |
+
**Default negative prompt**:
|
173 |
+
```
|
174 |
+
lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, artist name, displeasing
|
175 |
+
```
|
176 |
+
|
177 |
+
**Default configuration:**
|
178 |
+
|
179 |
+
Euler a with around 25-30 steps, CFG 5-7, and ENSD set to 31337. Sweetspot is around 28 steps and CFG 7.
|
180 |
+
|
181 |
+
## Training Configurations
|
182 |
+
|
183 |
+
- Finetuned from: [Animagine XL 3.1](https://huggingface.co/cagliostrolab/animagine-xl-3.1)
|
184 |
+
|
185 |
+
**Pretraining:**
|
186 |
+
|
187 |
+
- Dataset size: ~23,600 images
|
188 |
+
|
189 |
+
- GPU: 1xA100
|
190 |
+
|
191 |
+
- Optimizer: AdaFactor
|
192 |
+
|
193 |
+
- Unet Learning Rate: 3.75e-6
|
194 |
+
|
195 |
+
- Text Encoder Learning Rate: 1.875e-6
|
196 |
+
|
197 |
+
- Batch Size: 48
|
198 |
+
|
199 |
+
- Gradient Accumulation: 1
|
200 |
+
|
201 |
+
- Warmup steps: 100 steps
|
202 |
+
|
203 |
+
- Min SNR Gamma: 5
|
204 |
+
|
205 |
+
- Epoch: 10 (epoch 9 is used)
|
206 |
+
|
207 |
+
**Finetuning:**
|
208 |
+
|
209 |
+
- Dataset size: ~6,800 images
|
210 |
+
|
211 |
+
- GPU: 1xA100
|
212 |
+
|
213 |
+
- Optimizer: AdaFactor
|
214 |
+
|
215 |
+
- Unet Learning Rate: 2e-6
|
216 |
+
|
217 |
+
- Text Encoder Learning Rate: - (Train TE set to False)
|
218 |
+
|
219 |
+
- Batch Size: 48
|
220 |
+
|
221 |
+
- Gradient Accumulation: 1
|
222 |
+
|
223 |
+
- Warmup steps: 5%
|
224 |
+
|
225 |
+
- Min SNR Gamma: 5
|
226 |
+
|
227 |
+
- Epoch: 10
|
228 |
+
|
229 |
+
- Noise Offset: 0.0357
|
230 |
+
|
231 |
+
## Added Series
|
232 |
+
|
233 |
+
**Wuthering Waves**, **Zenless Zone Zero**, and **hololiveEN -Justice-** have been added to the model.
|
234 |
+
|
235 |
+
## Special Thanks
|
236 |
+
- **My co-workers(?) at CagliostroLab** for the insights and feedback.
|
237 |
+
|
238 |
+
- **Nur Hikari** and **Vanilla Latte** for quality control.
|
239 |
+
|
240 |
+
- **Linaqruf**, my tutor and role model in AI-generated images.
|
241 |
+
|
242 |
+
## License
|
243 |
+
**UrangDiffusion 1.2** falls under the **[Fair AI Public License 1.0-SD](https://freedevproject.org/faipl-1.0-sd/)** license.
|