How to run Stable Cascade?

[stableuser]

How to run Stable Cascade in Automatic1111 webui?

[CarbonatedPastries]

It doesn't support it at the moment. I was actually curious where they are thinking on support, since ComfyUI already has support.

[Soulreaver90]

When it comes to supporting the latest trends, A1111 wouldn't be the first choice. Comfy and SD.Next tend to implement new things sooner.

[CCpt5]

There's a "temporary" extension for A1111 that will allow you to generate w/ Stable Cascade (it an independent tab) here:

https://github.com/blue-pen5805/sdweb-easy-stablecascade-diffusers

Hopefully it's natively implemented sometime in the near future.

[beansfotos]

That didn't work for me either. I was looking for the same thing, some way to run SCascade in webUi

[rinaldop]

It is working for me on SD -Forge.

[GnatGoSplat]

Working fine for me on Automatic1111 dev branch. Takes about 3 minutes to generate an image on an RTX 3070 though.

[Niknew9]

I installed a separate version:https://github.com/EtienneDosSantos/stable-cascade-one-click-installer
RTX4060Ti 16Gb; 1 Image 1024x1024; Guidance Scale-7; Steps-30
2.69it/s-00:07s 2.21it/s-00:04s
1 image - 11s

[plxin]

For me, it can’t be used in the dev branch.

[springheeledjack66]

It is working for me on SD -Forge.

also running forge here, it shows up but does generate anything, I get no kernal found to luanch and then get a error, how are you getting it to work?

[enn-nafnlaus]

How do you even select it? It's a whole directory of .safetensors with their own subdirectories.

[GnatGoSplat]

It should have its own StableCascade tab.

[enn-nafnlaus]

Works, thanks

[CCpt5]

Working fine for me on Automatic1111 dev branch. Takes about 3 minutes to generate an image on an RTX 3070 though.

That's a very long time. I have a 4090 and it takes about the same time as SDXL to generate an image.

You should check in NVidia settings, under 3D settings, and see if the relatively new option for "Prefer Sysmem Offload" is toggled on (or device default). If it is your system may be moving some of the data for generation to CPU RAM while inferencing which can cause extremely performance drops. I've had this happen during training when I accidently had that toggled on.

[thethiny]

@GnatGoSplat I use the anime ones version 1 that are around 200 MB. Idk where you find 6GB models cuz the max I have right now are 1.6GB. Get them from civit ai

[GnatGoSplat]

@thethiny Ah, forgot about civit.ai. I got some of my checkpoints from there, but I've only been using SDXL and all of those are 6GB or bigger. Can you tell me the name of the 200MB one you have? They don't have a sort by size, I browsed a whole bunch of anime and cartoon ones, but couldn't find any below 1.99GB.

[thethiny]

@GnatGoSplat it's called Zelda breathe of The wild and the first version was small

[GnatGoSplat]

@thethiny I found that one, but it's been archived and no longer available for download. V1 shows it used to be 1.99GB, so I went and downloaded a random 1.99GB one which is the smallest I could find and switched that to the active checkpoint, but still run out of VRAM in Stable Cascade. I might be stuck with super slow Stable Cascade until it's made into a standard checkpoint that can run on 8GB VRAM.

[thethiny]

Well I guess you're out of luck. Just need to wait until official cascade support in that case. @GnatGoSplat

[Niknew9]

Why run a cascad on A1111? Install separately and everything is fine. https://github.com/EtienneDosSantos/stable-cascade-one-click-installer

[enn-nafnlaus]

It's a shame that there's no native support, only a poorly integrated plugin (which to be fair to the author, is much better than nothing!). Stable Cascade's main weakness seems to be the relatively poor diversity that it generates. But it handles large resolutions with low VRAM and is very pretty (as well as better at text than XL), so it'd be superb in img2img. But we can't do that.

[plxin]

That sadly does not show the tab for me at all... I hope it gets added soon to a1111.

I finally found out why this tab is not displayed. This plug-in depends on a package: diffusers. You can check the error log of your service.

[Mohsyn]

If anyone has successfully got Stable Cascade working on automatic 1111 or Forge, can you please share your directory structure of how models are organized to be accessed by sdweb-easy-stablecascade-diffusers