How should I fix this error?

[yurisukichan]

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument index in method wrapper__index_select)

[johnnybpena1989]

This just started happening to me as well. But only when I run the ControlNet extension.

[johnnybpena1989]

I Also get this error if I check the Low VRAM setting inside of the Extension:
RuntimeError: mat1 and mat2 shapes cannot be multiplied (154x1024 and 768x320)

[E36lewis]

make sure your not on v2 model, CN only works with 1.5 model atm

[Bodzijun]

I noticed that this happens to me when I use VAE together with a module that already has VAE built into it. If you turn it off in the settings, there is no error.

[Piscabo]

Where is this setting you refer to? Cannot find it in my WebUI settings. It happens when I use controlnet with newer 2.0 and 2.1 models. But others doing the same does not have this problem?

[Bodzijun]

Tab Setting -> Stabble Diffusion -> SD VAE.
For 2 and 2.1 SD it is better to disable VAE. For version 1.X, enable it, unless otherwise provided by the Model description.
This will work if you have a VAE file at all in the webui-Models-VAE folder. Otherwise, it may be located in the folder along with the Model of the same name.By default, they are not there at all, they are downloaded by you separately. Therefore, if you did not download them, then the problem may be completely different.

[Fmada]

how to check if model has vae (e.g. from training) - I need to vae from stability.ai to remove blue artifacts when using lora

[Murcielagoz99]

I believe this may be correct. I disabled VAE but nothing happened so I switched MODELS and ControlNet works again!

[MrDo1965]

Check the Low Vram option, in ControlNet, solves the problem, in local install and colab, for me.

[sledgehkg]

I'm getting this with controlnet off and none set in VAE. But only when I have a LORA in the prompt

[alexmihaic]

did u got a solution?

[alexmihaic]

RuntimeError: mat1 and mat2 shapes cannot be multiplied (77x3072 and 768x4)

This happens to me every time I want to use a LORA. Help?

[vt-idiot]

I am running into the same issue, but onyl with certain LoRA

[vt-idiot]

@alexmihaic downgrading to 1.4.0 made it go away. I don't give a rat's ass about SDXL, so I guess I'll stay at 1.4.0 until they figure out what they broke

git checkout 394ffa7b

[sjaak31367]

For me it's solved if I uncheck Settings -> Compatibility -> Lora/Networks: use old method
Running version v1.5.1 ( 68f336b )

[criptycal]

bro I finally find the solution, if you are using XL models just download openpose XL model in hugging face and paste inside controlnet models folder

[criptycal]

https://huggingface.co/lllyasviel/sd_control_collection/tree/main

[DGdev91]

It happens when you are trying to load sd 1.5 models while using SDXL.
For example a model for controlnet made for SD 1.5, or also a LORA