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)

[Skyfallen228]

When I try to use txt2image the first image is generated normally but when I try to generate the next one it shows "RuntimeError: CUDA out of memory. Tried to allocate 58.00 MiB (GPU 0; 4.00 GiB total capacity; 3.36 GiB already allocated; 0 bytes free; 3.44 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF" and after that, if I try to repeat the generation, it shows "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)"
My GPU is 1650 Super 4GB
Yes, I know that there is not enough video memory, but I use the parameters in webui-user.bat --precision full --no-half --medvram Because without them, I only generate completely black pictures.
Img2img works fine the only problem is with txt2img
How to fix it?

[vasanthwho]

did u find the solution??

[TrueTGamer]

did anybody solve it?

[TrueTGamer]

did u find the solution??

hey dude did you manage to fix it?

[Eprise1701e]

Having same issue with 1660. Worked on previous install, then went and updated and now cannot used txt2img.

[qvisionstudios]

Having the same problem. Was working fine with using the standard model. Once I switch to any another loaded model, I get the error. Running NVIDIA GeForce RTX 3090, 24GB.
Edit using SD v1.5

[Skyfallen228]

My friend told me to put this "set COMMANDLINE_ARGS=--precision full --no-half --lowvram --always-batch-cond-uncond --opt-split-attention" in webui.bat and it worked
I can now make pictures 896x896

did u find the solution??

[TrueTGamer]

My friend told me to put this "set COMMANDLINE_ARGS=--precision full --no-half --lowvram --always-batch-cond-uncond --opt-split-attention" in webui.bat and it worked I can now make pictures 896x896

did u find the solution??

Nope, i've tried this aswell and much to no good it hasn't worked at all. I moved over to midjourney and decided to subscribe to their service. Hopefully when i get a better pc i might have better luck i guess

[CapRogers9527]

My friend told me to put this "set COMMANDLINE_ARGS=--precision full --no-half --lowvram --always-batch-cond-uncond --opt-split-attention" in webui.bat and it worked I can now make pictures 896x896

did u find the solution??

Nope, i've tried this aswell and much to no good it hasn't worked at all. I moved over to midjourney and decided to subscribe to their service. Hopefully when i get a better pc i might have better luck i guess

after you put this "set COMMANDLINE_ARGS=--precision full --no-half --lowvram --always-batch-cond-uncond --opt-split-attention" in webui.bat
if there is still a error warning like this:"You can skip this check with --disable-safe-unpickle commandline argument."
the way to solve this problem is put "--disable-safe-unpickle" command in the bat file you start up your Stable diffusion
the code that line should be "python launch.py --disable-safe-unpickle" after done.

My friend told me to put this "set COMMANDLINE_ARGS=--precision full --no-half --lowvram --always-batch-cond-uncond --opt-split-attention" in webui.bat and it worked I can now make pictures 896x896

did u find the solution??

Nope, i've tried this aswell and much to no good it hasn't worked at all. I moved over to midjourney and decided to subscribe to their service. Hopefully when i get a better pc i might have better luck i guess

after you put this "set COMMANDLINE_ARGS=--precision full --no-half --lowvram --always-batch-cond-uncond --opt-split-attention" in webui.bat
if there is still a error warning like this:"You can skip this check with --disable-safe-unpickle commandline argument."
the way to solve this problem is put "--disable-safe-unpickle" command in the bat file you start up your Stable diffusion
the code that line should be "python launch.py --disable-safe-unpickle" after done.

[Moxie1776]

Mine started when I added the extension sd-webui-refiner, and went away when I removed it.

[suonnon]

Mine started when I added the extension sd-webui-refiner, and went away when I removed it.

Problem solved!!!! Thank you so much!! Sad to say goodbye to SDXL refiner integration when generating but happy to get rid of this damn issue!

[Wanninayake]

My friend told me to put this "set COMMANDLINE_ARGS=--precision full --no-half --lowvram --always-batch-cond-uncond --opt-split-attention" in webui.bat and it worked I can now make pictures 896x896

did u find the solution??

was able to fix the issue. thanks a lot

[marks202309]

Mine started when I added the extension sd-webui-refiner, and went away when I removed it.

Problem solved!!!! Thank you so much!! Sad to say goodbye to SDXL refiner integration when generating but happy to get rid of this damn issue!

Yeah, works after I disabled the SDXL refiner. Good findings.

[zzzgithub]

I am trying to make stylezed animateddiff animation.
Prompt: <lora:LCM Lora_V1.5:1>
Sampling: LCM
Controlnet: lineart
Controlnet1: Ipadaptor
Animatediff model:mm_sd_v15_v2.ckpt
when I am trying to generate shows error:
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 weight in method wrapper_CUDA___slow_conv2d_forward)
please help how can I solve the problem. Other solutions above did not help me.

[lynrayy]

Mine started when I added the extension sd-webui-refiner, and went away when I removed it.

THANKS

Fuck this refiner. It didn't work at all anyway.

[lynrayy]

Solution:
Remove or disable this shit

[Cheesdumpling1234]

A solution that worked for me was switching to the original Model.ckpt and making a new hypernetwork with that model running

[Cheesdumpling1234]

A solution that worked for me was switching to the original Model.ckpt and making a new hypernetwork with that model running