Any reason for using torch 1.13.1 instead of 2?
Any reason for using torch 1.13.1 instead of 2?
AYes, Torch version 2 does not work with the Dreambooth extension for training.
Ftorch 2 was not working with dreambooth
AEven the 2.0.1cu118 version is still not working.
TAh well that makes sense then. I was just looking at this https://github.com/d8ahazard/sd_dreambooth_extension/blob/main/D8-DreamBooth.ipynb but haven't tried it.
GitHub
Contribute to d8ahazard/sd_dreambooth_extension development by creating an account on GitHub.
AI am just busy looking into an issue with Jupyter notebook not starting with my template, but RunPod has a bug in launching templates where you can't set environment variables, so waiting for them to fix that before I can test further.
AHmm, I see he has
--torch2
CThanks for helping out! I did install C++ Build tools, and a restart fixed the issue. Now I am facing a new one No module named 'psutil'. Let me check if I can just type pip install...
CYes, it works! Thanks @nikgrid & @Furkan Gözükara SECourses
N@Casvseb Nice! you're welcome People have helped me..I'm just paying back 
FYes give it a try
Fi'm a bit afraid of trying/running SimSwap, it uses a lot of models from chinesse people xDD
Fit is possible to sandbox python apps or something like that?
Fmodels that are downloaded from mega, google drive, etc
Fbest way is run on google colab :d
Fbut no privacy for files
Fcorrect
Fyou can use runpod
Fi think it is private
Fsecure cloud + disk encryption should be
F@Furkan Gözükara SECourses any tip for the face swrapper picture?
Fportrait? zoom on face?
Fyou mean regarding latest tutorial?
Fit is 128px so limited
Fyes
Foh, 128px?
Fi'm doing it with a 512x512 image ! XD
Fand it seems to work!
Fit's swapping using cpu, so slow
Fbut if it works using cpu... then it's as easy as spawn a local VM and run the model there with privacy and no issues about malware
Fyes but quality is very low if you pay attention
Fit even works with full hd too :d
Fyes it uses cpu
Fyou can use vmware
Fbut... image is scaled 128px? or i need a 128px image?
For you mean the video is downscaled to 128px?
Fi think faces are downscaled to 128
Fthen replaced
Fthen upscaled
Fhere
F
Foh yeah, full screen is horrible resolution
Shey everyone, long time no see, i ordered a new computer with a 4090 so i'll be back to learn some more =]
Fdamn it, the script could say how many frames are processed from the total
Finstead of
FStatus: swapping in progress...
.....................................................................................................................................................................................................................................................................................................................................................................................................................................................................
.....................................................................................................................................................................................................................................................................................................................................................................................................................................................................
Swill i still have to use lora when training with a 24gb card?
Fno lol
Sgood i've only had good luck with the collab dreambooths
F@Furkan Gözükara SECourses i think the simswap method is very powerful because you can pre-train the model yourself!
