you need to have class in prompt instance
you need to have class in prompt instance
Fi cant tell much maybe your lora config incorrect
Fyou can train on google colab
Ffor free dreambooth
Z
Zdoes that look right?
Fwhat do you mean?
FI mean... you use "photo of" only in class prompt: "photo of man"
Fwhy not add "photo of" also to the instance prompt? why not "photo of ohwx" ? why only "ohwx"
Zstarting to come around at 26 epochs
Fyou dont provide classification folder you checked it generates those?
Fyou can add. all those tokens will also get trained with your images
Fit is like using image captions
Zi was using prior preservation but it never seemed to train the model, i went in the dreambooth chat and one of the mods said i didn't need to use class images to train
Fi made that test in this video
FYouTubeSECourses - Software Engineering Courses
Sign up RunPod: https://bit.ly/RunPodIO. Our Discord : https://discord.gg/HbqgGaZVmr. New best training settings for DreamBooth training in Automatic1111 Web UI. If I have been of assistance to you and you would like to show your support for my work, please consider becoming a patron on 
https://www.patreon.com/SECourses
Playlist of #StableDif...
https://www.patreon.com/SECoursesPlaylist of #StableDif...
Zwere you using lora or dreambooth tho? does it make a difference?
Fdreambooth
Fmake difference
Zi was using images generated by class, but it never seemeed to train the model correctly, their faces were either blurry, or looked nothing liek the subject
Zi was using lora extended with lion tho, i think it's buggy
Fi'm using image captions and the face fidelity is perfect. photos looks like real ones, but not much styling
Zcaptions like a woman standing in the park with green trees and mountains, like that?
Fyes my captions are like that
F"photo of xzy wearing...... standing in....."
Fi don't include "woman" in the captions
Fthe purpose matters
F8 epoch done i am gonna test model on my pc
Fwhile training goes on
Zdo you use BLIP or just do them yourself?
Fbtw i hope you can include a "conclussions" at the end of your videos
Fi tested captions. it was hard to get subject for me
Zwish we could just use an alpha mask and get rid of backgrounds completely
Fafter 1 hour of data i can't even remember and compare anymore XDD plus, i can't judge if you look similar or not
Zwhy even need captions?
Fso a conclussions like "EMA is good; XZY is probably bad" would be awesome
Fbut holy s., your videos are pure gold
Zmy worst fears have come, it turned my subject asian. That's why i had asian as negative prompt, her eyes are very asianish
Z
Zwell not my worst fears, but i suspected it would happen
Fin this video i used best settings so you can see them quickly @felixsanz https://youtu.be/m-UVVY_syP0
YouTubeSECourses - Software Engineering Courses
RunPod: https://bit.ly/RunPodIO. Discord : https://bit.ly/SECoursesDiscord. Get your own #Midjourney level model for free via Style Teaching. If I have been of assistance to you and you would like to show your support for my work, please consider becoming a patron on https://www.patreon.com/SECourses
Playlist of #StableDiffusion Tutorials, Au...
https://www.patreon.com/SECoursesPlaylist of #StableDiffusion Tutorials, Au...
Fyou cant give negative while training
Fi watched them all, that one too XD
F(several times)
F
Fyeah, the negative prompt is for sampling
Zi suppose negative is why my subject overtrained so quickly cause it had a lot less options to choose from
Zcan she come back from being asian? or will she be asian forever?
Fzono you have confusion with settings
Fyour class prompt is not used
Fif you dont use classification images during training