Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,28 @@
|
|
| 1 |
---
|
| 2 |
license: creativeml-openrail-m
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: creativeml-openrail-m
|
| 3 |
---
|
| 4 |
+
|
| 5 |
+
Prompt with "sarasa"
|
| 6 |
+
|
| 7 |
+
Training details:
|
| 8 |
+
|
| 9 |
+
- Trained with TheLastBen's fast-DreamBooth notebook
|
| 10 |
+
- data set: 111 concept images after data augmentation + a number of cusomized reg images (cannot remember)
|
| 11 |
+
- learning rate 2e-6 for 5000 steps
|
| 12 |
+
- text encoder rate 15%
|
| 13 |
+
|
| 14 |
+
Problems:
|
| 15 |
+
|
| 16 |
+
- The generated images can get blurred easily; this may be due to the use of anime screenshot
|
| 17 |
+
- Some cropping does not include face and this is not properly tagged, and therefore it may generate such images
|
| 18 |
+
- The pendant is never properly generated
|
| 19 |
+
|
| 20 |
+
Example generations (with inpainting):
|
| 21 |
+
|
| 22 |
+
|
| 23 |
+
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
|
| 27 |
+
|
| 28 |
+
|