K@Furkan Gözükara SECourses Would you mind me asking... I might have either missed it completely, missed it because of the flood of available info or I have not missed at all and there actually is no such video, BUT.... Finetuning... We have tutors for creating checkpoints/embeddings/LORA's from subjects/persons/styles, but where is the tutorials for creating complete finetuned checkpoints (Or embeddings/LORA's if possible). I mean like training them on thousands of images like some models are.
Most models on CivitAI seem to be just people mixing other models together and that feels more like a trial and error method that can turn out "how ever".
Most models on CivitAI seem to be just people mixing other models together and that feels more like a trial and error method that can turn out "how ever".
Fdmn it failed. looks like we need a bigger karma having account :d
KLet's say I want to train a model on all the different mushrooms of the world...
Fi dont have any fine tuning video
Fi plan but i didnt get any good training dataset to make yet
KGood (or bad) then I haven't missed it 
KAh I see
Fguys if you upvote this i would appreciate very much. meanwhile working on deep fake video : https://www.reddit.com/r/nvidia/comments/13vbej1/artificial_intelligence_breaks_new_ground_in/
reddit
0 votes and 4 comments so far on Reddit
Kseems a lot of manual work is required aswell then . Are there any plans to release those regularization set of images for free someday ?
Fty
Fguys can you check and tell me which sound is best? tom hanks : https://github.com/FurkanGozukara/Stable-Diffusion/blob/main/test.md
GitHub
Best Stable Diffusion and AI Tutorials, Guides, News, Tips and Tricks - Stable-Diffusion/test.md at main · FurkanGozukara/Stable-Diffusion
KI open the terminal in the folder stable diffusion webui like this /workspace/stable-diffusion-webui# git pull and it said I already got the lastest version
KAlso I try to install Kohya Dreambooth like in your new video but it won't let me train with "Gradient accumulate steps"
KI tried GAS 25-16-8-4-2 nothing work, the training is started only if I use GAS 1. I'm using Runpod RTX 3090 24GB
Fdont use GAS and see if it works
KYes I can trained with GAS 1 but if I increase it to 2 it won't run because of this error
KMany people on Reddit said that bigger Batch Size and bigger GAS will improve quality so I tried to test that but with RTX 3090 I can't even increase the GAS
Fit reduces quality for teaching a person
Fit is for speed up training process
KI would say 4640, 3520 or 960
F@Khang that font spacing is really annoying, it happens to me on firefox too. i use chrome just to avoid this shit XD
Fdoes batch size really affects quality?
AI think only 4640 and 3520, not 960.
KYeah I know. It's hard to read as hell. I'm using Chrome and still got this problem
KI'm not sure about other types of training, but in my style training (where I work with game characters in different forms like humans, humanoids, monsters, etc.), I noticed that increasing the batch size from 1 to 4 really makes a big difference in quality. So, I want to increase it even more. However, the RTX 3090 on RunPod can only handle a maximum batch size of 4.
KKinda liked 960 as well though
KIt had something
Fgo here
chrome://settings/fontsthat fixes it
Fbut improves it? Furkan just said that a bigger batch size reduces quality!
AYep, that was my understanding from his tutorials too.
AWhat is the recommended Python version for Dreambooth training again?
F3.10
AAny 3.10? or a specific point release?
Fany
AGreat, the RunPod template installs 3.10.10 but mine installed 3.10.6 so i was a bit concerned about that.
Fboth are ok
ASpot pods are saving me a lot of $ 
A
Fsoooo.. i've written a technical article about how stable diffusion works internally (but easy to understand for almost everyone)
Fthis is good to know because it helps understand better the most important parameters
FI appreciate if someone wants to read it and even more if give me feedback 
FUnderstand in a simple way how Stable Diffusion transforms a few words into a spectacular image.
Fyes
Fi think it reduces for teaching a person
Fthanks
Fi am currently focused on deep fake tutorial
F