[Feature Request]: SD XL support

[dorioku]

Is there an existing issue for this?

• [X] I have searched the existing issues and checked the recent builds/commits

What would your feature do ?

SD XL has released 0.9 model and ComfyUIhas supported two weeks ago, ComfyUI is not easy to use.

Proposed workflow

support SD XL

Additional information

No response

[wzgrx]

sdxl0.9Already leaked

https://huggingface.co/sayakpaul/controlnet-sdxl-circles

https://drive.google.com/file/d/1J-2KhUG7ZvcN6H_-BIoQgdhkShWRU2YZ/view?usp=sharing

[imacopypaster]

https://drive.google.com/file/d/1J-2KhUG7ZvcN6H_-BIoQgdhkShWRU2YZ/view?usp=sharing

lively link: https://openload.cc/Zeg57704zb/sdxl09_torrent

[catboxanon]

Duplicate of #9680

[449693691]

Webui will help a lot of people if it supports SDXL 0.9.

[Panchovix]

Also does works on fork branch of auto111 webui, on the diffusers branch.

[missionfloyd]

It's not really worth using in its current state anyway.

[altoiddealer]

I just want to comment that I can see the future a little ways out.

Everyone still uses Reddit for their SD news, and current news is that ComfyAI easily supports SDXL 0.9 out of the box, tutorial videos already available, etc.

Our favorite YouTubers everyone is following may soon be forced to publish videos on the new model, up and running in ComfyAI.

Just saying, this may be a pivotal moment (I hope not)

[freecoderwaifu]

Considering all the people who are still sore over the last time a model leaked and may or may not still be out to get this repo, it's better to wait until the proper and legal 1.0 model releases.

[Yorha4D]

In fact, ComfyUI is more stable than WEBUI(As shown in the figure, SDXL can be directly used in ComfyUI) @dorioku

[Panchovix]

Now, SDXL supports have been added to vladmandic automatic

[dorioku]

In fact, ComfyUI is more stable than WEBUI(As shown in the figure, SDXL can be directly used in ComfyUI) @dorioku

ye I know, i also use comfyui. but web-ui is easier to use and suitable for more one

[missionfloyd]

In fact, ComfyUI is more stable than WEBUI(As shown in the figure, SDXL can be directly used in ComfyUI) @dorioku

That's Blender.

[dorioku]

In fact, ComfyUI is more stable than WEBUI(As shown in the figure, SDXL can be directly used in ComfyUI) @dorioku

That's Blender.

it's comfyui for blender

[Rokkonrol]

Now, SDXL supports have been added to vladmandic automatic

Specifically the dev branch: https://github.com/vladmandic/automatic/tree/dev

Update notes on how to enable SDXL support for the dev branch: vladmandic@9f96d4f

EDIT: OUTDATED INFO/LINKS

[Panchovix]

Now, SDXL supports have been added to vladmandic automatic

Specifically the dev branch: https://github.com/vladmandic/automatic/tree/dev

Update notes on how to enable SDXL support for the dev branch: vladmandic@9f96d4f

Yep! Thanks for the aclarations. Have been using in the last hour and it is working pretty good.

[Yorha4D]

In fact, ComfyUI is more stable than WEBUI(As shown in the figure, SDXL can be directly used in ComfyUI) @dorioku

That's Blender.

it's comfyui for blender

In fact, we also developed the WEBUI version in November last year But due to many errors in the WEBUI backend that prevented stable parameter transfer, we temporarily suspended it (it was stopped in April 2023) I hope WEBUI can become more standardized and stable in the future

[johndpope]

use / watch this branch
https://github.com/vladmandic/automatic

[ImranBug]

SDXL have been made public officially as well now, it is there on stabilityai profile of hugging face: https://huggingface.co/stabilityai/stable-diffusion-xl-base-0.9 https://huggingface.co/stabilityai/stable-diffusion-xl-refiner-0.9

[PatrickJnr]

use / watch this branch vladmandic/automatic

I get size mismatch for model

[johndpope]

https://www.reddit.com/r/StableDiffusion/comments/14ub4zn/sdnext_vlad_with_sdxl_09_working_right_now/ https://www.reddit.com/r/StableDiffusion/comments/14uduq1/introducing_sdnexts_diffusion_backend_with_sdxl/ (personally yet to get it working with automatic. Works fine with comfyui. )

[ollobrains]

any updates on when core automatic1111 will support it. ?

[brennanerbz]

any update?

[dhwz]

you can try this extension https://github.com/lifeisboringsoprogramming/sd-webui-xldemo-txt2img till it gets fully implemented

[franzkruhm]

Ok probably a very newbie question, but what is preventing me from dropping the current SDXL file into the model directory of my current install? are tthere fundamental changes to the code to be implemented?

[mirek190]

SDXL has 2 models working as once.

[turkyden]

Fucking Awesome !

[dorioku]

Ok probably a very newbie question, but what is preventing me from dropping the current SDXL file into the model directory of my current install? are tthere fundamental changes to the code to be implemented?

It has different unet layer and double clip

[petedan]

Why don't we just use diffusers for XL? https://github.com/huggingface/diffusers/releases/tag/v0.18.1

[cerseinusantara]

interesting

[dorioku]

I see sdxl branch was created da464a3fb39ecc6ea7b22fe87271194480d8501c

[zhangvia]

use / watch this branch https://github.com/vladmandic/automatic

i still met mismatch error. although the webui will load it ignore the error, i get a completely black result image using stable xl

[dorioku]

i get a image by sdxl branch, master branch will support it soon, this issue can be close

[wzgrx]

i get a image by sdxl branch, master branch will support it soon, this issue can be close

Will there be a trimmed version of sdxl0.9 12gb? A model suitable for 6.8GB small graphics memory? It seems that the minimum graphics memory requirement for SDxl0.9 is starting at 12GB.

[altoiddealer]

SDXL 0.9 will be superceded by 1.0 very soon.

The sdxl branch seems to just add very preliminary support since it doesn’t use the refiner model (?)

[dhwz]

refiner is an img2img model so you've to use it there

[altoiddealer]

refiner is an img2img model so you've to use it there

ComfyUI allows processing the latent image through the refiner before it is rendered (like hires fix), which is closer to the intended usage than a separate img2img process… but one of the developers commented even that still is not the correct usage to produce images like those on Clipdrop, stability’s discord bots, etc