i'm using image captions and the face fidelity is perfect. photos looks like real ones, but not much
i'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
Fand negative of that class prompt is used
Zso if you don't use images, then the class prompt is completely ignored?
Fonly if dreambooth extension generates classification images
Zahhhh
Fbut classification images doesn't affect the subject right?
Fit just protects the "woman" class from dissapearing
Fso model doesn't forget about all other womans
Zwell every photo i've had has been a woman
Zjust the wrong kind of woman
Fprobably but not sure about exact implementation of the extension
Fbut that has nothing to do with class or negative prompts in sampling
Zi have like 2000 images i've already generated for photo of woman and photo of a beautiful woman
Finitial testing
F
Fi've already done img2img with videos
Flike extracting 8000 frames of pngs and applying img2img to them all XDD
Fbut not for styling, just to change face like a deepfake
Fi am working for a perfect video :d
Fand result is kinda decent
Fdo you have it anyhwere uploaded?