Software Engineering Courses (SECourses)•2y ago

Your results are very similar to mine. I wonder why these results are coming out and I want to find

Your 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?

GGen-AI Your results are very similar to mine. I wonder why these results are coming out...

J2_Nedo•7/11/24, 7:34 AM

if you invert the colors in photoshop, the colors become correct but the resolution is terrible

Kkareem9000 Can someone point me to the latest dreambooth patreon guide link

Furkan Gözükara SECourses•7/11/24, 9:57 AM

https://www.patreon.com/posts/very-best-for-of-89213064

Patreon

The Very Best Workflow For SDXL DreamBooth / Full Fine Tuning - Res...

Get more from SECourses: Tutorials, Guides, Resources, Training, MidJourney, Voice Clone, TTS, ChatGPT, GPT, LLM, Scripts by Furkan Gözükara on Patreon

Furkan Gözükara SECourses•7/11/24, 9:58 AM

and download and look at runpod_8400_Steps.png attachments

Furkan Gözükara SECourses•7/11/24, 9:58 AM

@Dazzastrous [4090]

Ppapanton any idea what would cause the trained model to spit out images like this?

Furkan Gözükara SECourses•7/11/24, 9:58 AM

config error

TTheToday If in our dataset we have images with different expressions and emotions, would ...

Furkan Gözükara SECourses•7/11/24, 9:58 AM

i would do 2 training and compare and see which one worked better

Furkan Gözükara SECourses•7/11/24, 9:59 AM

but i think ohwx man may work better than captioning still try it

MMaxivi looks like te2 is training at its full rate on the latest kohya. in case anyon...

Furkan Gözükara SECourses•7/11/24, 9:59 AM

yes this should be provided. in my config i provide it

DDarth Teethius Anyone kick the tires on the sd3 training on one trainer yet?

Furkan Gözükara SECourses•7/11/24, 9:59 AM

i am waiting merged into main still has errors

GGen-AI Your results are very similar to mine. I wonder why these results are coming out...

Furkan Gözükara SECourses•7/11/24, 10:00 AM

i trained 6 models like 2 days ago

Furkan Gözükara SECourses•7/11/24, 10:00 AM

worked perfect with my config.

FFurkan Gözükara SECourses i trained 6 models like 2 days ago

Gen-AIOP•7/11/24, 10:08 AM

I 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.

GGen-AI I just solved a problem I had. When CompyUI connected vae to vae decoder through...

Furkan Gözükara SECourses•7/11/24, 10:09 AM

makes sense because training werent broken

FFurkan Gözükara SECourses makes sense because training werent broken 😄

Gen-AIOP•7/11/24, 10:14 AM

I 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.

GGen-AI Your results are very similar to mine. I wonder why these results are coming out...

kareem9000•7/11/24, 10:22 AM

Spooky

GGen-AI I trained the model using your config file, but when I didn't connect the VAE lo...

Furkan Gözükara SECourses•7/11/24, 10:37 AM

use models in auto1111 and if they work directly then model trained accurately

Maxivi•7/11/24, 11:06 AM

@Furkan Gözükara SECourses have you tried trainings with sdxl noise offset of 0.0357 and the scheduled huber loss settings

Maxivi•7/11/24, 12:10 PM

@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

Théo•7/11/24, 2:09 PM

After 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)

TThéo After 30 training and 5 dataset retake to have the best one, i finally achieve t...

Maxivi•7/11/24, 2:11 PM

are you training on realvis4 and using the model on a different checkpoint?

Théo•7/11/24, 2:32 PM

traing on realvis4, using on realvis4

Théo•7/11/24, 2:33 PM

i 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

Maxivi•7/11/24, 3:12 PM

and all your images are cropped at 1024x1024?

Théo•7/11/24, 3:22 PM

ok 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

Maxivi•7/11/24, 3:23 PM

use adetailer with correct settings, and with hires fix use 4x ultrasharp or any other good upscalers

Théo•7/11/24, 3:24 PM

my problem was that i wasnt upscalling enough, 1.25 or 1.5 is not good enough for the adetailer to work properly

Théo•7/11/24, 3:25 PM

but 2x upscaling is good, now i have the (body proportion) problem but its easier to fix than not be able to inpaint

Maxivi•7/11/24, 4:50 PM

yeah i usually upscale with 1.75x to 2x with 60 steps on adetailer

MMaxivi <@205854764540362752> when you get some time can you try this prompt out on your...

Furkan Gözükara SECourses•7/11/24, 10:51 PM

some prompts will overwrite your likeliness

Furkan Gözükara SECourses•7/11/24, 10:51 PM

like super Mario

Furkan Gözükara SECourses•7/11/24, 10:51 PM

because the model will assosicate it with a certain man

Furkan Gözükara SECourses•7/11/24, 10:51 PM

super Mario - original dataset

Furkan Gözükara SECourses•7/11/24, 10:51 PM

or super man

Furkan Gözükara SECourses•7/11/24, 10:51 PM

it is like ohwx man as brad pitt same

TThéo After 30 training and 5 dataset retake to have the best one, i finally achieve t...

Furkan Gözükara SECourses•7/11/24, 10:52 PM

for mid and distant shots i use adetailer and it works

Furkan Gözükara SECourses•7/11/24, 10:52 PM

you wont get good face with those without adetailer

Furkan Gözükara SECourses•7/11/24, 10:52 PM

because the model inherent training rsolution is too low

FFurkan Gözükara SECourses some prompts will overwrite your likeliness

Maxivi•7/12/24, 1:12 AM

i noticed training a little bit of the text encoder 2 helps with that

Maxivi•7/12/24, 9:07 AM

@Furkan Gözükara SECourses what is mulit line captioning? the wildcard? what does it do and how do you use it

Khalifa•7/12/24, 12:13 PM

@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?

MMaxivi i noticed training a little bit of the text encoder 2 helps with that

Furkan Gözükara SECourses•7/12/24, 1:41 PM

Nice tips

MMaxivi <@205854764540362752> what is mulit line captioning? the wildcard? what does it ...

Furkan Gözükara SECourses•7/12/24, 1:41 PM

Just write a prompt each line

Furkan Gözükara SECourses•7/12/24, 1:41 PM

It will randomly use during training each one

KKhalifa <@205854764540362752> do you have any guidelines/guide on how to pick images? a...

Furkan Gözükara SECourses•7/12/24, 1:41 PM

I explain in tutorials but I don't have a specific tutorial yet

Furkan Gözükara SECourses•7/12/24, 1:42 PM

I am composing a good dataset to make a dedicated turorial

Théo•7/12/24, 4:43 PM

i 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

smoke•7/12/24, 8:13 PM

Does 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..

daroth•7/12/24, 8:16 PM

what do you mean? theres hundreds of trained good stuff on civitai

smoke•7/12/24, 8:18 PM

Yeah of course but preferably from people here with advice and that can share training parameters

smoke•7/12/24, 8:18 PM

Civitai is full of good stuff ofc