After done training I need to download swarmui to generate images right? also what exactly are regul
After done training I need to download swarmui to generate images right? also what exactly are regularization images for?
Freg images not useful for flux i tested
Fyou can use swarm ui forge comfy
Fi prefer swarm
Xhttps://youtu.be/HKX8_F1Er_w?feature=shared
I will follow this video after training is down just installing part and lora part
I will follow this video after training is down just installing part and lora part
YouTubeSECourses
Do not skip any part of this tutorial to master how to use Stable Diffusion 3 (SD3) with the most advanced generative AI open source APP SwarmUI. Automatic1111 SD Web UI or Fooocus are not supporting the #SD3 yet. Therefore, I am starting to make tutorials for SwarmUI as well. #StableSwarmUI is officially developed by the StabilityAI and your m...
Fyep great one
Lso does it not matter if the training images are kinda "high res" like 4000px instead of 1024 on the wide side?
Fit will auto downscale
Fto your train res
Dnot available 
LI now have unified the aspect ratios -- so now all pics are one of the following aspect ratio: 0.5, 0.666, 0.8, 1, 1.5, 2
I let it train over night and we´ll see -- also I add some more images for training
I let it train over night and we´ll see -- also I add some more images for training
Fpick alt config but at least get 2 GPUs so to get more ram
F@NicB we need more gpus 
D96GB RAM, will it be enough?
DCoupon code is not valid for this GPU Type and/or Quantity. Please try again.
Fyep greqat
Fonly up to 4 :/
Fteesting ipndm
F
Fwow ipndm makes it much more realistic
D@Dr. Furkan Gözükara in my case instance Prompt is the keyword trigger?
but in class prompt what do I write? it is not a person
but in class prompt what do I write? it is not a person
DI will try a first test using:
Rank_1_FAST_4x_GPU
Rank_1_FAST_4x_GPU
Fuse joy captions and like chatgpt
Fto get your images described
Ffrom there get idea
Flike i used 3d rendering
Frank 1 fast is good
Flol i publish thousands of dollars worth article, lots of research info and still some people complains like why i use lora rank 128 instead of 32 you can use if you wish
you can use if you wish DAccording to chatgpt it tells me that my image is a “Countryball”.
but using some caption models trained in replicate or huggingface it tells me it's a “cartoon illustration”.
but using some caption models trained in replicate or huggingface it tells me it's a “cartoon illustration”.
Ftry both on flux
D
Fand see which is more accurate generating
Fcartoon illustration looks more accurate
Dis it necessary to describe that it is in 2D?
DI have a doubt, and excuse my ignorance....
but when using the prompt is it necessary to write “cbznft cartoon illustration”? i.e. use both?
or am I wrong?
but when using the prompt is it necessary to write “cbznft cartoon illustration”? i.e. use both?
or am I wrong?
Lstupid question: should or can training images also contain things to learn which are rare? like 98% of all the training images of the same person have blue eyes -- and 2% have green eyes (like they wearing colored contact lenses)?
Lanother example -- how the person looks like from "behind"
Fthey should contain images
Fonly you want as output
Fafter training
Fi feel like this new sampler beter
D@Dr. Furkan Gözükara I am going to try without *.txt captions, 800 images 1024x1024, what epoch should I use? I think there is a formula, and how often the checkpoint?
DSave every N epochs?
D
Dthat's the tricky part, understanding how much to use epoch, train batch size.
Fsave every 1 epoch
Ftrain batch size max 2 brings some speed up
Fon rtx a6000
Depoch?