but result was not good ☹️
but result was not good 
Fi never got good results with Prodigy
Ftry bf16 too
Fbut reason could be 768x768
Falso add this parameter to optimizer : --max_grad_norm=0.0
Fi will do another research for LoRA
Pok i will follow these tips for next training
J: A tensor with all NaNs was produced in VAE. This could
be because there's not enough precision to represent the picture. Try adding --no-half-vae commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check. How to fix?
be because there's not enough precision to represent the picture. Try adding --no-half-vae commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check. How to fix?
SI wonder what is quality difference if I don't use classification images? training would be much faster, but would I lose a lot of quality?
MI used this parameters for Prodigy, and trained SD 1.5 base model: "use_bias_correction=True" "weight_decay=0.5" "decouple=True" "betas=(0.9, 0.99)" "d_coef=2" "safeguard_warmup=False"
Fthis happens when model totally overtrained
JI don't use with model lora I use model checkpoint Every model doesn't work.Basicly sd 1.5 I found this problem.
Sso is SDXL training essentially the same as 1.5 training just using an SDXL model? @Dr. Furkan Gözükara
Fthe parameters changes too
Flearning rate
F
Sso not 0.000001
Fyes
FI am searching best parameters for SDXL DreamBooth for weeks now
FHuge Stable Diffusion XL (SDXL) Text Encoder DreamBooth training comparison
U-NET is always trained
PNG info (prompts) are in alts
All images are 1024x1024 so download full sizes
https://twitter.com/GozukaraFurkan/status/1711345323900100643
U-NET is always trained
PNG info (prompts) are in alts
All images are 1024x1024 so download full sizes
https://twitter.com/GozukaraFurkan/status/1711345323900100643
Bv23.10.0 adetailer got released. Just noticed because now in the preview of the image a red box appears around the detected face
Bgood job, from a first glance it does not look like it is changing that much? what is your impression
F2023-10-07
- v23.10.0
- 허깅페이스 모델을 다운로드 실패했을 때, 계속 다운로드를 시도하지 않음
- img2img에서 img2img단계를 건너뛰는 기능 추가
- live preview에서 감지 단계를 보여줌 (PR #352)
Fi think it still adds some more extra details
Fso 2e-6 looks like nice. 3e-6 also nice but a little bit starting to cook
Bhow did it influence the training time?
Bah you already wrote in patreon it is slower
Fye a little bit more slower
Flike 1.5 second / it to 1.75 second / it
SThank you so much to @Dr. Furkan Gözükara and @Sandstormleader after several hours of training and solving some errors the model came as I expected this is definitely a game changer process
MI have now made a perfect Lora with Prodigy, but you have to use Clip Skip 2 when generating and then it will be flexible.
Finteresting never tested clip skip 2. this is sd 1.5?
MYes, I trained with RealisticVision 5.1, and I generated the images in the same model.
MIf I generated on clip skip 1, the model was inflexible, but on clip skip 2 everything was fine.
And I thought right: I trained it with clip skip 2 and then clip skip 1 can be generated well.
And I thought right: I trained it with clip skip 2 and then clip skip 1 can be generated well.
MSo now the SDXL training works properly with TE, but what's up with 1.5 dreambooth?
Rwhat is the best/fastest gpu deal on Runpod to do inference on Auto1111? A100 or A6000? I find A6000 very slow, slower than 4900...
Fyes but using kohya
Fi think sd 1.5 dreambooth also should work but need good hyper parameter testing. also kohya missing EMA 
F4090 could be fastest with price ratio
Rok, yea, that's what i thought. a6000 takes twice as long to infer
Kuse "runpodctl" to transfer big Data between your desktop and runpod https://github.com/runpod/runpodctl
GitHub
| RunPod CLI for pod management. Contribute to runpod/runpodctl development by creating an account on GitHub.
| RunPod CLI for pod management. Contribute to runpod/runpodctl development by creating an account on GitHub.
TI tested 4090 with Kohya on Windows 11 and it was 1.89 it. On Windows 10 it is 1.06 it. Seems like there is a huge driver issue with W11
MI'm speaking about kohyaSS 1.5 dreambooth training. Can I train SD1.5 with TE?
Fuploading hugging face is even faster
Fit works too slow on windows
Fdriver issue
Fyes we found you can
Fbut it is inferior
TBut I'm using it on Windows 10 and it works really fast. Seems like Windows 11 is the problem
