madebyollin
commited on
Commit
•
7a6d5a0
1
Parent(s):
499a66e
Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,30 @@
|
|
1 |
---
|
2 |
license: mit
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: mit
|
3 |
+
library_name: diffusers
|
4 |
---
|
5 |
+
|
6 |
+
# `stage-a-ft-hq`
|
7 |
+
|
8 |
+
`stage-a-ft-hq` is a version of [Würstchen](https://huggingface.co/warp-ai/wuerstchen)'s **Stage A** that was finetuned to generate sharper details and textures.
|
9 |
+
`stage-a-ft-hq` works with any Würstchen-derived model (including [Stable Cascade](https://huggingface.co/stabilityai/stable-cascade)).
|
10 |
+
|
11 |
+
> TODO: comparison goes here
|
12 |
+
|
13 |
+
## 🧨 Diffusers Usage
|
14 |
+
|
15 |
+
```py
|
16 |
+
# TODO: code goes here
|
17 |
+
```
|
18 |
+
|
19 |
+
## Explanation
|
20 |
+
|
21 |
+
Image generators like Würstchen and Stable Cascade create images via a multi-stage process.
|
22 |
+
Stage A is the ultimate stage, responsible for rendering out full-resolution human-interpretable images.
|
23 |
+
|
24 |
+
The original Stage A tends to render smoothed-out textures with a distinctive grain pattern on top.
|
25 |
+
|
26 |
+
`stage-a-ft-hq` was finetuned on a high-quality dataset in order to generate cleaner, sharper, more realistic textures with fewer distinctive artifacts.
|
27 |
+
|
28 |
+
## Recommended Settings
|
29 |
+
|
30 |
+
To generate highly detailed images, you probably want to use `stage-a-ft-hq` (which improves very fine detail) in combination with a large Stage B step count (which improves mid-level detail).
|