the inpaint would be using the subject LoRA only. Right. Ok
the inpaint would be using the subject LoRA only. Right. Ok
Fi got to sleep see you tomorrow hopefully
Fso many videos i need to make so little time :d
Rthank you so much! Get your well deserved sleep Dr
RSo are you simply using your Face LoRA InPainted?
Pstill a lot of testing to do. at least I didn't turn into a robot or something halfway through the video. style and structure remained consistent throughout. so it's progress.
Pwicked awesome results
KLook awesome 
CIf all my Dreambooth checkpoint outputs are grainy, what does this indicate?
CAll outputs on txt2img with the trained model are grainy as well. Running 23GB Vram.
C
CDoes this have to do with the VAE?
Pcould be your model is overbaked
Ptoo much training
Pthis image def looks overbaked
KThat's a super cool idea, Dr. Furkan! Can we also use it for checking styles and objects? We're kinda stuck because SD can churn out way more images than we can check manually. It doesn't have to be perfect, but if it could rank each image from 1 to 10, it would save us a ton of time and effort
CAt all sample steps, training images are blurry:
C500, 600, 700
C10 High quality training images with diverse lighting and angles, all cropped to 512
Pyou should run a x/y plot with your checkpoints
CWill do one sec
CX/Y is not working for some reason but here is 1000 steps:
C
C800 steps
C
C600 steps
C
C@PhillieD Does the quality look off to you as well?
P600looks best
CDo your training sample images look as grainy as mine?
Pdon't quite understand what you're asking
CI feel like I can be achieving much higher quality, feels like there is no-vae distortion.
CThese are some of my outputs from the dreambooth/samples folder
CAll the sample images look like this, at every ckpt
Pso you haven't tried generating new images using the checkpoints you created? these are just the images generated during training?
CAll those "forest guardian" images are new generations at each checkpoint
CThey still feel very low res
PI mean theyre low-res because its 512x512. a lot depends on the samples, denoising, cgf, and then upscaling. hard to tell really
CI could also be hallucinating for staring at Auto for 4 hours lol
CHere are a couple examples of my model merged with Lyriel and 2.1 VAE
C
Plooks good to me
KHi guys, my SD only shows grid images. How can I make it show both grid images and all the other images that it has generated?
Ahi guys what is the best model to train dreambooth realistic faces on? thanks
PYouTubeSECourses
#Kohya SS web GUI DreamBooth #LoRA training full tutorial. You don't need technical knowledge to follow this tutorial. In this tutorial I have explained how to generate professional photo studio quality portrait / self images for free with Stable Diffusion training.
Our Discord server
https://bit.ly/SECoursesDiscord
If I have been of assis...
Our Discord server
https://bit.ly/SECoursesDiscord
If I have been of assis...
Fthe method currently i am working on works only at faces because it will compare similarity of generated face with your training data face
Fis this happening after latest update? could be a bug
Frealistic vision atm
Ci got kandinsky running locally
Cresults are pretty impressive for an untuned model