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

Black image as result #144

Open
IDelant opened this issue May 7, 2024 · 5 comments
Open

Black image as result #144

IDelant opened this issue May 7, 2024 · 5 comments

Comments

@IDelant
Copy link

IDelant commented May 7, 2024

I`m trying to generate any image, but in results get blank black image. I follow the guide, starting with fresh install, and this appears.
Already tried:
Reinstall from auto installer and from manual download
Reinstall to another ssd
Using different checkpoint model
Updating gpu drivers

Notes:
Sometimes it's generate an image, but only in between steps, like with 20 steps image appears in 5 10 and 15, but on final step is black.
Also, all tries of generating image larger than 320x320 produces black image from first step.
Using Samples or Itteration with walue greater than 1 also results in black image
Using different model create similar results. Appears on steps, but does't in finals
Earlier i`ve installed RealESRGAN, and in results getting black images cause of wrong pytorch version
Screenshot_1

Screenshot_2
Screenshot_3

@Temder
Copy link

Temder commented May 8, 2024

The only thing I can think of that produces black images is the nsfw filter. Do you have it turned off?

@IDelant
Copy link
Author

IDelant commented May 8, 2024

The only thing I can think of that produces black images is the nsfw filter. Do you have it turned off?

I tried different combinations with safety checker/precision/caching/unloading but unfortunatly getting same results

@painebenjamin
Copy link
Owner

The other situation in which a black image can occur is when a number is out of range - for instance, it gets converted to RGB where one of those values is less than 0 or greater than 255. That occurs due to floating-point rounding errors in some less-stable models; for instance, SDXL 1.0 was infamous for its rounding errors when operating in half-precision, so much so that the standard still is not to use SDXL 1.0's VAE and instead to use SDXL 0.9's.

Could you please let me know what hardware you are using? This has occurred with several other users, and I'm trying to find a commonality between you all so I might be able to fix. At the moment, I'm still unable to reproduce this issue.

A few possible options for remediation:

  1. Under System > Settings, there is an option labeled Precision Mode which defaults to Use Half-Precision When Available. Please try setting this to the alternative option, Always Use Full Precision.
  2. Using the More Model Configuration button will allow you to configure a specific VAE to use. There are two drop-in SD 1.5 VAE's that you can select instead of using the default baked VAE; please try selecting either of them (sd-vae-ft-ema or sd-vae-ft-mse) to see if that changes the final result.

Please let me know if these fix it for you so I can hopefully better address this if it happens in the future, too. Thank you!

@IDelant
Copy link
Author

IDelant commented May 10, 2024

Could you please let me know what hardware

GTX 1660 Super 6gb
Ryzen 5 3600
16gb ram
Fresh win 10 on which i immediatly downloaded realesgan and dedmucs, day later enfugue

At the moment, I'm still unable to reproduce this issue.

Maybe at some point i can give you remote access so you can possibly debug and catch this

Precision Mode

I tried different combinations with safety checker/precision/caching/unloading but unfortunatly getting same results

Tried, did't help

More Model Configuration button will allow you to configure a specific VAE to use. There are two drop-in SD 1.5 VAE's that you can select instead of using the default baked VAE; please try selecting either of them (sd-vae-ft-ema or sd-vae-ft-mse) to see if that changes the final result.

Please let me know if these fix it for you so I can hopefully better address this if it happens in the future, too. Thank you!

Tried all vae's in list + clearvae from civit al, with both half and full precisions, didn't help

Updated notes:
Tried SD webui and encountered an error with vae and net. Both produced NaN values in half precision modes, but using full precision helped get images.

i'm not an expert in AI at all and it's just my thought. but did preview steps decode using selected vae or some integrated and fast solution? Like in SD web you can select vae for preview decoding and maybe preview decodes in full precision, but result in half, so result gets NaN values and image are black.

Also sometimes getting log error(or not), but doesn't exactly know what causes it

stderr (may not be an error:) DEBUG:root:Initializing MLIR with module: _site_initialize_0
DEBUG:root:Registering dialects from initializer <module 'jaxlib.mlir._mlir_libs._site_initialize_0' from 'E:\\Enfugue\\enfugue-server\\jaxlib\\mlir\\_mlir_libs\\_site_initialize_0.pyd'>

DEBUG:jax._src.path:etils.epath was not found. Using pathlib for file I/O.

WARNING:xformers:A matching Triton is not available, some optimizations will not be enabled.

Error caught was: No module named 'triton'

Sorry if i wrote this with a grammar mistakes, i' not a native speaker and was too lazy to check spelling with translator.
Thanks!

@xXEyeDropXx
Copy link

Hello !

Same issue here..
My hardware :

GTX 1660 Super 6gb
Ryzen 5 1600AF
16Gb RAM

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

4 participants