are you able to train with regular model?
are you able to train with regular model?
Jno weird
Fyou must have a config or install error
JI install a fresh version later.
I have the hunch that torch: 2.0.0 is the probem
I have the hunch that torch: 2.0.0 is the probem
Jthanks
Zany general recommended settings for lora extended with lion to train a human?
Zright now i'm doing 21 mimages, lora unet is 5e-7, 1.8e-5 tenc, min learning rate 1e-9 150 epochs
Zbatch and grad size is 1
Fi think try without torch 2 to see if works or not
Fi still didnt have time to test lora yet :/
SI tried lycrois but never had good results mate
Zyeah mine barely looks like the subject
Chello here impossible to make appear drambooth in SD diffusion after your video, i put the good quot in "requirement versions" after that i have delete"venv" files.
Cbut nothing more appear
Cif somone have a idea 
thank you
thank you
Fhello. do you get any error? can you show screenshots
Cmessage.txt5.45KB
Cthis is m log to the webui
C
Fok best way is
Fdelete venv
Fthen
Finstall requirements of dreambooth
Fas shown here
Falso dont forget to add --xformers to your command line arguments
Ffor patreon supporters i personally install and setup on their computer via google remote and discord voice chat
Fif you cant make it
Ci got a 3070 with 8gb do i need xformers
Cso i delete again venv, relunch webui and do your commands ?
Dfor dreambooth videos with niche topics such as getting a midjourney aesthetic, i think faster videos where you show the end result generated image at the beginning so that people know what the tutorial will teach them and then going step by step in an ordered sequence on how to achieve that end result are best. You already have plenty of long detailed videos on dreambooth so if someone wanted a more in-depth understanding of its settings you can recommend people watch those first at the beginning of a video.
Cthank a lot for your help doctor 
DAlso, here are some of the images generated using your last custom dreambooth style video. Thank you for all your work!
Cafteryour manipulationi'v got that message : ERROR: Could not install packages due to an OSError: [WinError 5] Accès refusé: 'D:\AI\stable-diffusion-webui\venv\Lib\site-packages\google\~rotobuf\internal\_api_implementation.cp310-win_amd64.pyd'
Check the permissions.
Check the permissions.
Cmessage.txt20.87KB
Cnothing more whn i relaunch webui
GDreambooth extention seems to be broken. I can't get it to show on the tab in stable-difussion anymore
Fyes you need xformers
Fso this was done good or bad in that sense? thank you for feedback
Fwow success
Fby the way i will also upload trained model
Fpermission error
Fcan you restart pc and try again
Cof course
Fso not fixed?
Ci gonna restart in 2 minutes
Cnew error message
message.txt20.97KB
Cbut drambooth is appear
Cahaha thank you so much Doc
Dyes! though for better flow in the video I would've moved the first three minutes ( specifically when you're talking about the training dataset and classification images) over to eight minutes into the video when you start talking about the style training settings to introduce the concepts, and instead begin the video with the steps for training yourself. Because I'm still a novice, I assumed we'd also be generating the dataset for the model, so I was confused why we never got to that part at first 
. I also see that there's people complaining about over-explanation in the comments, it could perhaps be helpful to skip dialogue about trial and error and just talk about what worked, such as when you talked about the training dataset being 1024x1024. It'd be better if you instead mentioned in passing that larger resolutions gave bad results and that the viewer should use 512x512 images. The video was still very informative despite this! Thank you for your work!
. I also see that there's people complaining about over-explanation in the comments, it could perhaps be helpful to skip dialogue about trial and error and just talk about what worked, such as when you talked about the training dataset being 1024x1024. It'd be better if you instead mentioned in passing that larger resolutions gave bad results and that the viewer should use 512x512 images. The video was still very informative despite this! Thank you for your work!
HAnyone check out the Nvidia press release today?