which cuda version is best for dreambooth rn?
which cuda version is best for dreambooth rn?
Fyou can make venv and install things there
Fbest way
Flatest works
Flatest cuda and xformers work
TNot having luck with any, you are on 12.1?
Tcuda
HI solved the issue installing to venv works but it's not a permanent solution having python path setup correctly in system i think is a more permanent solution
Tchanging [optimizer] from AdamW8bit to AdamW solved my issue
HWhat do those reducing vram commands actually do?
HI mean i have vram available i just want it to use the right amount
FYouTubeSECourses
How to install Python, have multiple Python installations, set system wide default Python version. How to compose venv for any Python installation, change Python default path, and install SD web UI properly. Discord: https://bit.ly/SECoursesDiscord.
If I have been of assistance to you and you would like to show your support for my work, please...
If I have been of assistance to you and you would like to show your support for my work, please...
Fwhat was your issue
Fuse them only if you get out of memory error
TError no kernel image is available for execution on the device at line 167 in file D:\ai\tool\bitsandbytes\csrc\ops.cu
HWhat do they actually do though? Like does it just push it to cpu or what?
HDoes it slow generations down?
Ehi, so i followed some of the tutorials and was able to manage to train stable diffusion on myself and got some safetensors from it to use. however, I'm wondering if i can somehow manage to integrate this model with some civit.ai anime model to create anime versions of myself using the civit anime model? Or is providing my own style images the only way to do so after? Sorry kinda new to this whole stable diffusion thing.
Tfrom nvidia documentation "For best performance, the recommended configuration is cuDNN 8.9.0 on H100 with CUDA 12.0, and cuDNN 8.9.0 on all other GPUs with CUDA 11.8, because this is the configuration that was used for tuning heuristics." seems like cuda 11.8 is best for now unless you have the h100
TThese support matrices provide a look into the supported versions of the OS, NVIDIA CUDA, the CUDA driver, and the hardware for the NVIDIA cuDNN 8.9.0 release.
DI wanted a trained file to play around with and see it's photorealism before deep diving into training it myself. Do you have any trained models? I can't see any links to trained models on your method
Fyes workflow is same
Finstead of base model use anime model
Eah perfect, thank you. that is easy
FDI have used realistic vision before I was curious about a trained model which used realistic vision as a base, such as your face one.
Fit is not different
Fonly contains my face
DYeah, okay if you have a model that you can share of yourself or any other person that you've done using that method I would appreciate that
Fsorry i cant share also i dont see any reason
DI'm just trying to prompt engineer and see what all a trained model using your method can do. If you can't share your face that's fine, perhaps you have done another face?
Fi have to record the new video but 1:36 am will get a hard coffe :d
Fi can train your face if you wish send me training data
DYes that would be amazing
DOkay I'll get it sent over in an hour
Fok
THaving some success now! hitting 2.5 it/s without 8 bit
Zfor some reason, i can train everyone but my wife, her pictures never look like her
Frecorded sorting video
Fhopefully will be published tomorrow. scripts shared on patreon
ISorry if ask again but anyone have a free colab with multicontrol net? won't destroy my poor 3060ti doing batch img2img 
AHi, does anyone have a workflow suggestion for replacing someone's face in a video? I couldn't get whole videos to look good so now I crop the video to a 512x512 square with the face mostly centered and then put it back in the full, uncropped video.
This is my process:
-set full uncropped video to 24fps in After Effects
- face track the video in AE and crop video to 512x512, face as big as possible
- Extract all frames to pngs
-remove every other frame (fewer imgs to process, will interpolate later)
- batch replace face with extension in img2tmg to give me a jump start
- img2img that batch with HED controlnet, original frames in the CN
- import that batch as a png sequence in AE at 12fps
-export new 512x512 video and run it into Flowframes to interpolate/increase fps
-import flowframes video into AE, place on the face in the uncropped original video
-make a simple mask to hide the edge of the blurred video
-make another layer of the video on top, 50% opacity, delay 1 frame
any suggestions? I'm using automatic1111 if that matters.
This is my process:
-set full uncropped video to 24fps in After Effects
- face track the video in AE and crop video to 512x512, face as big as possible
- Extract all frames to pngs
-remove every other frame (fewer imgs to process, will interpolate later)
- batch replace face with extension in img2tmg to give me a jump start
- img2img that batch with HED controlnet, original frames in the CN
- import that batch as a png sequence in AE at 12fps
-export new 512x512 video and run it into Flowframes to interpolate/increase fps
-import flowframes video into AE, place on the face in the uncropped original video
-make a simple mask to hide the edge of the blurred video
-make another layer of the video on top, 50% opacity, delay 1 frame
any suggestions? I'm using automatic1111 if that matters.
Htraining a model for style. dataset 100images. classification dataset 2500images， how many epochs do you suggest ?
H@Furkan Gözükara SECourses
XHey!! I'm absolutely new to ai art
I don't know a thing about it
How do I get started?
I don't know a thing about it
How do I get started?
Fi think you can do masked replace but i haven't done yet
F100 images is pretty high number. so i say train up to 100 epochs and save every 5 epoch to compare later
Fi have full guides
Ffirst start with python installation crucial