Add Stable Diffusion XL 1.0 Base Model support

divamgupta / diffusionbee-stable-diffusion-ui

Diffusion Bee is the easiest way to run Stable Diffusion locally on your M1 Mac. Comes with a one-click installer. No dependencies or technical knowledge needed.

https://diffusionbee.com

GNU Affero General Public License v3.0

12.11k stars 596 forks source link

Add Stable Diffusion XL 1.0 Base Model support #459

Open smurat opened 11 months ago

smurat commented 11 months ago

Stability AI released Stable Diffusion XL 1.0 Base Model.

I downloaded safetensors model file. But when i try to import model to Diffusionbee, an error occured. I couldnt find the error log location, sorry for that.

It would be awesome DiffusionBee support the Stable Diffusion XL 1.0 Based Models

everythingability commented 11 months ago

I just tried that too.

freaky-m0 commented 11 months ago

Woodentables commented 11 months ago

The divide by zero error happens in the for loop in line 44 of convert_model.py, but I don't have enough expertise to understand how to fix this. Could be that the model architectures are too different

cmosetick commented 10 months ago

Agree that a fix for this would be nice. I noticed that I can not import any FP32 model, it's not specific to SDXL 1.0. I can only import FP16 models. (This is on the 2.0 pre-release of diffusion bee)

DatanoiseTV commented 10 months ago

Error convert_model.py:48: RuntimeWarning: divide by zero encountered in divide
Traceback (most recent call last):
File "diffusionbee_backend.py", line 325, in <module>
File "convert_model.py", line 80, in convert_model
ValueError: The model is not supported. Please make sure it is a valid SD 1.4/1.5/2.1 .ckpt/safetensor file
[23883] Failed to execute script 'diffusionbee_backend' due to unhandled exception!

pmarreck commented 9 months ago

Encountered the exact same issue- tried to import a Stable Diffusion XL .ckpt file, got the same error.

Is anyone actually at the wheel, here? You can actually get SDXL running on macOS via the "Draw Things" iOS app (since iOS apps are now installable on Mac) and that works great... although I prefer the DiffusionBee UI

pmarreck commented 9 months ago

FWIW, the "coreml-stable-diffusion-xl-base-ios_split_einsum_compiled" model also doesn't work (perhaps because it's in some coreML-focused format that DiffusionBee doesn't recognize)

joshandersn commented 8 months ago

Just tried this and we still get the same error, any update on this?

smurat commented 8 months ago

I think that Mr. Divam Gupta has abandoned the project. While he isn't obligated to do so, it would have been very considerate of him to at least make an announcement about the project and his current personal situation. This way, he could inform us why he hasn't provided any updates regarding the project. Many people have a lot of respect for him and his project, particularly for their ability to run Stable Diffusion on their Macs smoothly and with great performance

pmarreck commented 8 months ago

The app "Draw Things" will get you there, but this app is noticeably more Mac-like. Ah well

DatanoiseTV commented 8 months ago

The app "Draw Things" will get you there, but this app is noticeably more Mac-like. Ah well

But it is much slower with an RX 580 on MacOS.

pmarreck commented 8 months ago

But it is much slower with an RX 580 on MacOS

Isn't that a 6 year old GPU that is as old as the Nvidia GTX 1080 of the same era, which it was also significantly inferior to?

https://gpu.userbenchmark.com/Compare/Nvidia-GTX-1080-vs-AMD-RX-580/3603vs3923

mattpilott commented 7 months ago

I get a similar error when importing the juggernaut xl model https://civitai.com/models/133005/juggernaut-xl