Anyone kick the tires on the sd3 training on one trainer yet?
Anyone kick the tires on the sd3 training on one trainer yet?
Mim pretty sure, patreon is pretty confusing, which they could organize it better
GYour results are very similar to mine. I wonder why these results are coming out and I want to find out how to solve them. Does anyone else have similar errors like this?
Jif you invert the colors in photoshop, the colors become correct but the resolution is terrible
FPatreon
Get more from SECourses: Tutorials, Guides, Resources, Training, MidJourney, Voice Clone, TTS, ChatGPT, GPT, LLM, Scripts by Furkan Gözükara on Patreon
Fand download and look at runpod_8400_Steps.png attachments
FFconfig error
Fi would do 2 training and compare and see which one worked better
Fbut i think ohwx man may work better than captioning still try it
Fyes this should be provided. in my config i provide it
Fi am waiting merged into main still has errors
Fi trained 6 models like 2 days ago
Fworked perfect with my config.
GI just solved a problem I had. When CompyUI connected vae to vae decoder through the vae loader node, the image was created normally. I was using the config file you shared with Patreon.
Fmakes sense because training werent broken 
GI trained the model using your config file, but when I didn't connect the VAE loader, the images came out the same as before. Is it possible that the model didn't train properly? I used the config file you shared on Patreon and trained with normalized images.
KSpooky
Fuse models in auto1111 and if they work directly then model trained accurately
M@Furkan Gözükara SECourses have you tried trainings with sdxl noise offset of 0.0357 and the scheduled huber loss settings
M@Furkan Gözükara SECourses when you get some time can you try this prompt out on your training?
Postive - cinematic photo Portrait of ohwx man as super Mario, red hat, fantasy, highly detailed, digital painting, artstation, concept art, sharp focus, illustration, art by Tony Sart and artgerm and randy vargas. 35mm photograph, film, bokeh, professional, 4k, highly detailed
Negative - drawing, painting, crayon, sketch, graphite, impressionist, noisy, blurry, soft, deformed, ugly, 3d render
DPM ++2sa Karras
1024x1024
CFG - 7
can you check if you get good likness
Postive - cinematic photo Portrait of ohwx man as super Mario, red hat, fantasy, highly detailed, digital painting, artstation, concept art, sharp focus, illustration, art by Tony Sart and artgerm and randy vargas. 35mm photograph, film, bokeh, professional, 4k, highly detailed
Negative - drawing, painting, crayon, sketch, graphite, impressionist, noisy, blurry, soft, deformed, ugly, 3d render
DPM ++2sa Karras
1024x1024
CFG - 7
can you check if you get good likness
TAfter 30 training and 5 dataset retake to have the best one, i finally achieve the - p e r f e c t - result thanks to @Furkan Gözükara SECourses however, mid-distance or distance shot are Totally destroyed, my dataset is perfect, used cinema grade camera, used cropper and resizer, tried mask training, adetailer, changing the number of step only for face, changing the denoising strengh, i can only make p r e f e c t close-up, selfie or whatever, but when the subject is in midrange like your picture in civtai, or fullbody, it's totally crap, even the face is not really "human" more like destroyed, any idea ? thanks for everything (also tried with and without reg)
Mare you training on realvis4 and using the model on a different checkpoint?
Ttraing on realvis4, using on realvis4
Ti have to admit i spent the last weeks understanding every little things possible and even myself cannot recognize a (fake) picture from myself, ,even my family, but im limited to close-up
Mand all your images are cropped at 1024x1024?
Tok i understood, to be possible to inpaint in a far away face, you need to Hires a lot, so the Adetailer can actually see the face to inpaint
Muse adetailer with correct settings, and with hires fix use 4x ultrasharp or any other good upscalers
Tmy problem was that i wasnt upscalling enough, 1.25 or 1.5 is not good enough for the adetailer to work properly
Tbut 2x upscaling is good, now i have the (body proportion) problem but its easier to fix than not be able to inpaint
Myeah i usually upscale with 1.75x to 2x with 60 steps on adetailer
Fsome prompts will overwrite your likeliness
Flike super Mario
Fbecause the model will assosicate it with a certain man
Fsuper Mario - original dataset
For super man
Fit is like ohwx man as brad pitt same
Ffor mid and distant shots i use adetailer and it works
Fyou wont get good face with those without adetailer
Fbecause the model inherent training rsolution is too low
Mi noticed training a little bit of the text encoder 2 helps with that
M@Furkan Gözükara SECourses what is mulit line captioning? the wildcard? what does it do and how do you use it
K@Furkan Gözükara SECourses do you have any guidelines/guide on how to pick images? as in what type of images should be picked for training?
FNice tips
FJust write a prompt each line
FIt will randomly use during training each one
FI explain in tutorials but I don't have a specific tutorial yet
FI am composing a good dataset to make a dedicated turorial
Ti tried different dataset style, always perfect quality with different clothing and background
i found out that it is not the best idea to add different facial expression since when using a prompt like "surprised"
if the model use an already surprised expression from the dataset it would be SUPER exaggerated and very weird, even using moderate prompt like "slightly"
I suppose that it would not be an issue if every picture from dataset were captioned, but i havnt found a good tutorial about dataset/captioning and method to use
Also i found that not be able to modify face is a pretty sad, adding a cyborg eye, or facepaint, makup, blood, dirt whatever, tried a lot of prompt even in the adetailer face parameter
I also found out that if your dataset is always with a sharp focused talent and blurry backround, whatever prompt you use, it will always be the same in Stablediffusion, always a sharp focused person, but blury background (i may doing something wrong), i feel like the "from single text file" method is pretty good for something quick, but if you want to achieve God level of training, you need to do the Super anoying long version of data training method
i found out that it is not the best idea to add different facial expression since when using a prompt like "surprised"
if the model use an already surprised expression from the dataset it would be SUPER exaggerated and very weird, even using moderate prompt like "slightly"
I suppose that it would not be an issue if every picture from dataset were captioned, but i havnt found a good tutorial about dataset/captioning and method to use
Also i found that not be able to modify face is a pretty sad, adding a cyborg eye, or facepaint, makup, blood, dirt whatever, tried a lot of prompt even in the adetailer face parameter
I also found out that if your dataset is always with a sharp focused talent and blurry backround, whatever prompt you use, it will always be the same in Stablediffusion, always a sharp focused person, but blury background (i may doing something wrong), i feel like the "from single text file" method is pretty good for something quick, but if you want to achieve God level of training, you need to do the Super anoying long version of data training method
SDoes anyone have some good results with SDXL trainings and could possibly share results? Mainly for close-ups / portraits.
Tried 1.5 but not too great for me..
Tried 1.5 but not too great for me..
Dwhat do you mean? theres hundreds of trained good stuff on civitai