guess mode

This commit is contained in:
lvmin
2023-02-20 06:35:11 -08:00
parent febbaac464
commit 9bd4929873

View File

@@ -248,7 +248,7 @@ Without prompt, the HED seems good at generating images look like paintings when
![p](github_page/uc6.png)
As you can see below, because **non-prompted mode is extremely difficult**, even the very deep ControlNet may also generate failure cases. (In the example below, the right-bottom one seems ok while others looks meh.) Because of this, the non-prompted mode seems a nice metric to test the capability of different encoder-based methods to control stable diffusion.
As you can see below, because **non-prompted mode is extremely difficult**, even the very deep ControlNet may also generate failure cases. (In the example below, the right-bottom one seems ok while others look meh.) Because of this, the non-prompted mode seems a nice metric to test the capability of different encoder-based methods to control stable diffusion.
![p](github_page/uc7.png)