Software Engineering Courses (SECourses)•2y ago

@Dr. Furkan Gözükara

crapthingsOP•4/5/24, 5:05 AM

crapthingsOP•4/5/24, 5:05 AM

crapthingsOP•4/5/24, 5:07 AM

this break cache progress i have to redo the cache again

Tuanny•4/5/24, 5:44 AM

what settings do i adjust in supir so that it doesnt hallucinate as much. i did fidelity and v0-F

Kallamamran•4/5/24, 7:23 AM

@Dr. Furkan Gözükara Is it possible to add a checkbox or something to make it possible for CLIP Interrogator to batch process subfolders as well?!

Kkareem9000 so technically i would try to run the dataset by supir before i start training o...

Neo•4/5/24, 8:00 AM

Yes, I am doing that now on most smaller resolution datasets for concepts, however I would be very wary of any noticeable artificial enhancements it may be doing to your dataset as that "artificialness" may be enhanced in your SD outputs when you include the LoRA in your inference. I find that keeping the prompt text influence fairly low ~4 or less seems to keep the upscale/detail recovery adequate without introducing artificialness into the AI upscaled image. Your mileage may vary...

MMorpheus SX343 Hi all, has anyone yet tried to be more accurate while training specific body p...

Neo•4/5/24, 8:26 AM

Not sure who has responded but I train LoRAs (presently using Kohya, will attempt to transition to OneTrainer soon) that require a high amount of quality and accuracy due to celebrity reproduction. What I am finding is that approaching model sized datasets (~300 images) is the key to increased accuracy across the board of your concept. This is the easiest and most straight forward method, but not always available if you can't acquire a dataset with enough good quality representations of your concept(s). So alternatives:
1) Train multiple concepts - This from reports is a mixed bag. I've done it for a TV celebrity who plays an iconic character with OK success. If I did it again, I would increase the size of my dataset for each concept to ~250 or more images.
Further to this, you will also need to watch out when training lots of closeups to get it to learn a concept as it will lose reference with your MAIN subject (depending on what your concept). So say your concept is a male celebrity but you want the abs more accurate, doing all closeups of the abs may cause it to lose reference to the main subject you want. After all it is giving you what its learned (just abs closeups) from the trigger word you associated it with.
2) Train with color association or some other novel machine learning method - See Civitai article here: (sorry it appears he's removed the article and model). There is a reddit post still here about it: https://www.reddit.com/r/StableDiffusion/comments/1aolvxz/instructive_training_for_complex_concepts/?share_id=TyIZDSSNoYSqsqZVR0SzW
3) Train them as separate LoRAs, this would operate like 1) but remove the pitfalls people claim happens with multi-concept training in a single LoRA.
4) Learn how to train a model for ADetailer specific for your subject (tbh, I have no idea where to start with this one, but it may be what you and I both need to get body part accuracy)

From the StableDiffusion community on Reddit: Instructive training ...

Explore this post and more from the StableDiffusion community

NNeo Not sure who has responded but I train LoRAs (presently using Kohya, will attemp...

Neo•4/5/24, 8:29 AM

@Morpheus A quick update: regarding training a custom ADetailer model: I did find this but don't presently have time to read it through: https://civitai.com/articles/1224/training-a-custom-adetailer-model

Yyunusos @Dr. Furkan Gözükara , is there any LoRA model we can play with that shows the f...

Neo•4/5/24, 8:31 AM

There is a Civitai article if I can find it that talks about manipulating the layers for LoRA optimisation, way over my head. I'll see if I can find it. here we go, not sure if it will help, but good luck LOL: https://civitai.com/articles/76/bdsqlsz-lora-training-advanced-tutorial1lora-block-training

bdsqlsz LoRA training Advanced Tutorial(1)：LoRA Block Training | Ci...

What is Lora block training? Lora Block training involves modifying the corresponding block weights during training through U-Net processing. First...

TTuanny what settings do i adjust in supir so that it doesnt hallucinate as much. i did...

Neo•4/5/24, 8:40 AM

I'm having great success with the text guidance set to 4 or lower to remove artificial influence from the base text prompt descriptions (or negative prompt).

NNeo Yes, I am doing that now on most smaller resolution datasets for concepts, howev...

kareem9000•4/5/24, 10:12 AM

Ahh I get you

Morpheus SX343•4/5/24, 10:36 AM

@Neo Thanks for all your answers. Good question if SD looses track of the concept or not when labeling the data correctly. I dont know if the trainer will understand a caption file like "abs of ohwx man" or not. For us humans it's clear, but not sure if the modell will understand to output those abs whenever I prompt a "naked body of ohwx man", that the abs will be part of it... Probably needed to be checked

Morpheus SX343•4/5/24, 10:39 AM

BTW guys, question of questions, everybody asks himself:
Any way to get better at avoiding being detected as AI image when using AI image detectors like HIV?

Morpheus SX343•4/5/24, 10:39 AM

HIVE

MMorpheus SX343 @Neo Thanks for all your answers. Good question if SD looses track of the concep...

Neo•4/5/24, 11:43 AM

Yeah if the base model has no idea what "abs" is, you will need to train it from scratch which means variation is king. You will need to do cropped shots, full shots, but always make sure the concept is present. Not sure how long it would take for it to understand abs if its not there in the base model you train off. I'd first run some inference checks or caption checks on the sourced model if you can for "abs" or "abdominals" etc.

Kallamamran•4/5/24, 11:47 AM

"photo of owhx man, abs, fit, abdominal muscles" and so on... That should suffice... If the model know abs it can generate ohwx man with abs.... I think!

NNeo Yeah if the base model has no idea what "abs" is, you will need to train it from...

Morpheus SX343•4/5/24, 12:05 PM

Yes but I hope the base model knows the basic concepts of body parts. I mean I'm not talking about special terms like "Musculus biceps brachii"

More like Legs, Knees, Abs, Arms, Chest, Breast, Back, Lower back, Shoulders... That's it. Very "generic" terms

Morpheus SX343•4/5/24, 12:07 PM

Thing is: My model has always exaggerated the breasts of my girlfriend or my own biceps and arm sizes... I wanna kind of correct that with some more detailled shots. Let's see if it'll work out

Morpheus SX343•4/5/24, 12:08 PM

I guess the base model only knows DDD extra large boobs

But I dont want these silicone boobs in my instance model

crapthingsOP•4/5/24, 12:47 PM

is this overfitting?

MMorpheus SX343 I guess the base model only knows DDD extra large boobs 😄 But I dont want these...

Kallamamran•4/5/24, 12:56 PM

That's the problem

Original message was deleted

Lelics•4/5/24, 2:45 PM

fake scam

NNeo Not sure who has responded but I train LoRAs (presently using Kohya, will attemp...

magical•4/5/24, 6:37 PM

the color coding is very interesting if it works it's amazing really. funny that the dude did all this work just to generate female genitals

magical•4/5/24, 7:09 PM

so does onetrainer masked training do this as well? can you mask parts of an image and describe them individually?

magical•4/5/24, 7:11 PM

hmmm if it is like in this example you can't color code it but only can mask one part of an image https://i.imgur.com/rqouPi9.png

Imgur

magical•4/5/24, 7:12 PM

so you'd have to mask the same image multiple times but that wouldn't work either because it wouldn't learn everything's relation to each other

Ccrapthings @Dr. Furkan Gözükara

Furkan Gözükara SECourses•4/5/24, 7:34 PM

ye please remove some images causes problem some times

TTuanny what settings do i adjust in supir so that it doesnt hallucinate as much. i did...

Furkan Gözükara SECourses•4/5/24, 7:35 PM

reduce text cfg

KKallamamran @Dr. Furkan Gözükara Is it possible to add a checkbox or something to make it po...

Furkan Gözükara SECourses•4/5/24, 7:42 PM

we already have that

Furkan Gözükara SECourses•4/5/24, 7:42 PM

which script are you using?

ada0436•4/5/24, 7:51 PM

Onetrainer question: Have 83 images - in TRAINING tab do I change Epochs 200 to 830? also what do I put under Stop Training After tab? 830? Thx

ada0436•4/5/24, 7:52 PM

used tier1_15.4GB fast scipt - RTX 4080

NNeo I'm having great success with the text guidance set to 4 or lower to remove arti...

Furkan Gözükara SECourses•4/5/24, 8:00 PM

good tips

NAMU•4/5/24, 8:00 PM

@Dr. Furkan Gözükara I'm watching your OneTrainer presets tutorial vid, and in the video it shows tier2_SDXL, but in this post (https://www.patreon.com/posts/96028218) it only has downloads for tier1. Just want to make sure I'm downloading the latest preset configs.

Patreon

OneTrainer Stable Diffusion XL (SDXL) Fine Tuning Best Presets | SE...

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