hello i cant generate beautiful anime women

[teamolhuang]

Did you like the remote VAE solution?

Issue Description

keeps saying 404 for remote VAEs
is it only me or

Version Platform Description

00:07:32-832996 INFO Logger: file="E:\sd-next\automatic\sdnext.log" level=INFO host="..." size=15384103
mode=append
00:07:32-836107 INFO Python: version=3.11.9 platform=Windows bin="E:\sd-next\automatic\venv\Scripts\python.exe"
venv="E:\sd-next\automatic\venv"
00:07:33-007343 INFO Version: app=sd.next updated=2025-04-14 hash=c2f2f5da branch=master
url=https://github.com/vladmandic/automatic/tree/master ui=dev
00:07:34-073611 INFO Repository latest available c6da1850614ea55a16ef02d45b51e57652211ecd 2025-04-16T13:25:42Z
00:07:34-080869 INFO Platform: arch=AMD64 cpu=Intel64 Family 6 Model 165 Stepping 5, GenuineIntel system=Windows
release=Windows-10-10.0.22631-SP0 python=3.11.9 locale=('Chinese (Traditional)_Taiwan', '932')
docker=False
00:07:34-082900 INFO Args: ['--use-zluda', '--use-nightly']
00:07:34-123500 INFO ROCm: AMD toolkit detected
00:07:34-611471 INFO ROCm: agents=['gfx1030']
00:07:34-612482 INFO ROCm: version=6.2, using agent gfx1030
00:07:34-613494 INFO Python: version=3.11.9 platform=Windows bin="E:\sd-next\automatic\venv\Scripts\python.exe"
venv="E:\sd-next\automatic\venv"

Relevant log output

00:10:21-360941 ERROR    Decode: type="remote" model=sdxl code=404 shape=torch.Size([1, 4, 135, 240])
                         url="https://x2dmsqunjd6k9prw.us-east-1.aws.endpoints.huggingface.cloud" args={'input_tensor_type':
                         'binary', 'shape': [1, 4, 135, 240], 'dtype': 'float16', 'partial_postprocess': False, 'output_type'
:
                         'pt', 'output_tensor_type': 'binary', 'scaling_factor':
0.13025, 'shift_factor': None} headers={'Date':

Backend

Diffusers

UI

ModernUI

Branch

Master

Model

Other

Acknowledgements

• I have read the above and searched for existing issues
• I confirm that this is classified correctly and its not an extension issue

What can be improved about the current solution?

make it work

What other VAEs you would like to see if the pilot goes well?

idk

Notify the members of the team

No response

[nitinmukesh]

is it only me or

Nope it's gone for now. Let's have word from team if this will be supported in future.

BTW nice title ;)

[teamolhuang]

Does hugging face or any other site have paid service that let me do remote VAEs?
Shit just crashes my PC every time I try to Full VAE. Shouldn't ve bought amd, these guys suck
I need my anime titties. please hugging face

[nitinmukesh]

Try using vae tiling and/or vae slicing.
Which model are you using for inference?

[iDeNoh]

Honestly I just wish they would address whether this is an error or if they plan on fixing it. I get that the circumstances are possibly muddy given that other post but a lot of people loved this feature, it's a shame to see it go and not know if it's coming back.

[yiyixuxu]

we’re working on a plan and will get back to you next week

[teamolhuang]

Thank you daddy

[teamolhuang]

bruh what are we drawing here

[iDeNoh]

Has there been any movement here? I've noticed that the remote vae for SD1.5 works as expected however sdxl and other models don't seem to work as expected as shown by the previous image

[teamolhuang]

i have since moved on to using tiled VAE locally instead. Some tips for later people if they couldn't run it:

1. My gig needed 40~50GB to run one, so I am using a 16GB VRAM card, and increased my DRAM up to 64GB, no more crash now. It even doesn't crash anymore without tile VAE. RAMs are pretty cheap nowadays, so try go local if you can.
2. sd-next provides something called "Texture Tiling" under VAE options - that seems a different thing, and made my anime women multiple heads. Use tiled VAE in the configs page instead.

yay anime womn

[yiyixuxu]

I think the endpoints are up running and we can close this issue no, no?
cc @DN6

[iDeNoh]

I think the endpoints are up running and we can close this issue no, no? cc @DN6

The endpoints seems to be up, but the sdxl vae is broken, is it possible the wrong vae is hosted on that endpoint? This image looked fine until the decode step.