Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Inaccurate test results, 80% of the time will be black image #7

Open
jaysunxiao opened this issue Jul 6, 2023 · 6 comments
Open

Inaccurate test results, 80% of the time will be black image #7

jaysunxiao opened this issue Jul 6, 2023 · 6 comments

Comments

@jaysunxiao
Copy link

Inaccurate test results, 80% of the time will be black image

@2575044704
Copy link

Same here

@2575044704
Copy link

same here, whatever models I use

@3Diva
Copy link

3Diva commented Sep 3, 2023

Yeah, I love the concept of this but it's FAR too sensitive and blocks quite a few images that are completely fine. I'm watching the images generate and they're fully clothed and fine but then at the last second the filter kicks in and turns the images completely black. It's blocking images that are nowhere near NSFW territory. While I Love the idea of this extension, it's far too sensitive and appears to block images that are completely normal and fine.

I have an older and slower computer and so it takes a couple of minutes for each image to generate - it's hugely frustrating to watch an image being generated and it's looking good and I'm excited to have it finish only to at the last second get turned into nothing but black.

While I would really love this feature if it wasn't so sensitive, the fact that it's blocking out images that seem to be fully clothed and fine is a bit frustrating, so sadly I'm going to have to disable this extension. I hope that it's able to be fine tuned a bit more and maybe updated to become less sensitive. :)

@jaysunxiao
Copy link
Author

really hope to support it

@kostia-official
Copy link

It's possible to make adjustments to it's sensitivity in this file:
https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion/safety_checker.py#L65

With some monkey patching of this file you can change adjustment value. But it should be really low, adjustment = -0.008 works better on images that I tested.

@MasterDenis
Copy link

It's possible to make adjustments to it's sensitivity in this file: https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion/safety_checker.py#L65

With some monkey patching of this file you can change adjustment value. But it should be really low, adjustment = -0.008 works better on images that I tested.

Hey how do we get this out of the box? I want to implement this better adjustment in my project, where do I find this file in my installation of A1111 in order to edit the value?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

5 participants