how to roll back a commit to a stable version in focus
how to roll back a commit to a stable version in focus
Fyou probably should do a fresh install
Nforge*
Fsame
Fclone into a new folder
Fthen find stable commit
Fopen cmd and git checkout
F6 months ago commit is this one
Fgit checkout b59deaa382bf5c968419eff4559f7d06fc0e76e7
Fand then install
Pcongrats for all hard work, supported you on patreon ! keep the good work
Nwhere can I see the commits?
Fawesome thank you so much
FGitHub
Contribute to lllyasviel/stable-diffusion-webui-forge development by creating an account on GitHub.
PI'll look at the full video tonight with some popcorn 
Is there a part for style/concept training, or is it mostly for character ?
Is there a part for style/concept training, or is it mostly for character ?Ffor style training i got a good dataset
Fwill make a new video for it and share results
Fbut watching this super important
Ji still didnt startet to scan my original paintings from the 1990, but now i will because of Flux, yes a style/concept video would be interessting too.
PI just made one from 1700s Japanese wood block prints and it came out amazing!!!!
P
P@Dr. Furkan Gözükara is that a big error or i can move forward with it ?
Pif you dont use xfromers - ignore
Pok thanks. I indeed done use it in comfy or forge
P
PBut if i have this, i should install the cu124 ? @protector131090
D@Dr. Furkan Gözükara I still have the bleeding problem, is imposible to train two or more people, may be is because the lr of TE and unet are incorrect, all bleeds, there is something missing, the base model can differentiate between "Michael Jackson" and "Michael Keaton" both man both Michael and the model can differentiate so is something fundamental that is missing, something did to train the base model that is missing now
J@Dr. Furkan Gözükara Hi, I successfully full finetuned flux but now I got diffusion_pytorch_model-00001-of-00003.safetensors, diffusion_pytorch_model-00002-of-00003.safetensors and diffusion_pytorch_model-00003-of-00003.safetensors after the training, I guess at this point I will have to fusion/merge theses 3 parts for created the whole checkpoint usable for Forge ( so reshard theses 3 shards ) . Would like to know a way to do it, thanks
�He claims to have finetuned Flux @Dr. Furkan Gözükara I thought it wasn't even possible yet? 
Jit's honestly easy with runpod, I think for now I need a tool or a script for "reshards " the 3 parts I get from the training
�Was it through Dreambooth tab in Kohya?
JI tried with kohya but switch to ostris ai tool kit
�Very interesting
Fit is because of cuda torch versions
Fyou can ignore xformers we dont use it
Ffor flux
Fsomeone said he trained 2 person perfectly with no bleed
Fuse 5 image person a 5 person b and 5 person a and b in same image
Fdid you follow my latest advice? no class token
Fwhat softwrae did you use? kohya doesnt do that
Ffine tuning fully possible
Fi am gonna do with kohya
�Oh ok it's becasue I haven't really seen anyone yet with an actual working model. I look forward to your research in that area though 
Fye it is next research
Fwith 8 gpus :d
Fi expect to go as low as 12 gb
Fbut we will see
Fthere are some serious optimizations
Dyes I did, no class, now I balanced better the dataset, and I will make a test with regularization images from the class person because I have different ages and genders, and let's see, my dataset was a little unbalanced all man looked like me even I didn't used any class
