It will use the default ckpt. Runpod has 2 default CKPT.
It will use the default ckpt. Runpod has 2 default CKPT.
AYes thats how you set your model
Cthank you
ZHey everyone, trying to get roop to work in linux, when i run python run.py --gpu , i get File "inswapper_128.onnx" does not exist"
Zi've already installed onnxruntime and onnxruntime-gpu
GThe old stable diffusion video instructions get this error now:
"Detected that PyTorch and torchvision were compiled with different CUDA versions. PyTorch has CUDA Version=11.7 and torchvision has CUDA Version=11.8. Please reinstall the torchvision that matches your PyTorch install."
Where do we get the matching torchvision?
"Detected that PyTorch and torchvision were compiled with different CUDA versions. PyTorch has CUDA Version=11.7 and torchvision has CUDA Version=11.8. Please reinstall the torchvision that matches your PyTorch install."
Where do we get the matching torchvision?
ATorch 1.13.1 is not available for 118, but 117 works on 118. Torch 2 is available for 118 but Dreambooth still does not work with Torch version 2.
AI assume you are using the RunPod 8.0.1 image?
GBut that is exactly how I got the error. In the past I used that exact code "pip3..." and Dreambooth extension worked in SD.
GYes exactly
AI recommend using the 6.0.1 image. 8.0.1 still seems buggy, Jupyter Lab doesn't work properly with the 8.0.1 image.
GHow do we find that? What is the template called?
A8.0.1 fixed a lot of bugs with 8.0.0 but still seems buggy.
GI'm using the one called "RunPod Stable Diffusion v1.5+v2"
Aimage name:
templae id:
templae id:
AThat should still be 6.0.1 and not 8.0.1.
AIt was 8.0.0 but they rolled it back because it was too confusing.
GIt is "runpod/stable-diffusion:web-automatic-6.0.1" and I still get that error...
GThere is not even an option to use 8.0 from what I see
Athe commands work with that image, last I checked
AYeah you don't wanna use that one yet.
ADid you activate your venv before installing torch?
GNo. That's not mentioned in the video
AIf you type what does it say?
G/workspace/venv/bin/python3
AThen its already activated
GAnd now SD is loading for some random reason 
GVery odd. First time I ever saw that 'not matching' error come up.
AYeah I've never seen it
PIn your opinion, what was the parameters in controlnet to achieve this result ? https://www.reddit.com/r/StableDiffusion/comments/13vp8sz/yes_my_4_year_old_sons_original_drawing_has_its/
Would like to try it on children drawings too !
Would like to try it on children drawings too !
reddit
1,451 votes and 160 comments so far on Reddit
Fhello. are you using latest automatic1111? this was fixed
Fyou should do pip uninstall torch torchvision torchaudio then reinstall
Fjust scribble and prompt may work
AThey resolved it by updating to the latest version 
ASo far in my limited testing I am seeing better results using than optimizer, but the dev branch of Dreambooth requires way too much VRAM.
Finteresting i wonder what caused the spike
AApparently not tested with Linux 
Arrow — Today at 10:42 PM
I haven't ran this with linux, so I don't really know. You'd probably have to search the web for the issue and tinker around. A quick thing you could try is disabling torch dynamo at the top of dreambooth/train_dreambooth.py:inner_loop
https://pytorch.org/docs/stable/_dynamo.html#torch._dynamo.disable
Arrow — Today at 10:42 PM
I haven't ran this with linux, so I don't really know. You'd probably have to search the web for the issue and tinker around. A quick thing you could try is disabling torch dynamo at the top of dreambooth/train_dreambooth.py:inner_loop
https://pytorch.org/docs/stable/_dynamo.html#torch._dynamo.disable
ZI've gotten it to create the images and upscale, but at the end ffmpeg does it's split, i get an error
unrecognized option 'crf' .
unrecognized option 'crf' .
message.txt2.72KB
Fi use davinci resolve
Fto reconstruct
Ffree edition
Zi'll see if i can install that on manjaro
Zso is it safe to say that the built in reconstruction using the gpu option is broken at this point?
Fgpu working for windows
Fon colab working as well
Zyeah idk why it won't reconstruct it on mine, followed all the reports that I can see, unfortunately they closed the discussion group
Zyou know me, i'll figure it out, will just take some time
Fyep you are really good
F@Dr. Furkan Gözükara did you change the tutorial for face swap?
Fwhere is the -f and -o parameters?
