so if there's a series of portrait style images, it will expand the canvas to make it square, and I
so if there's a series of portrait style images, it will expand the canvas to make it square, and I guess fill it with something?
Fif that image
Fhas person in distant
Fso that canvas can be expanded
Fif not it will crop
Fi think i added expand with white background too
N
NShouldn't larger VRAM bring more speed? The A800 (using Rank 1) has more than 3 times the VRAM of the 4090D, but the speed is barely increased. Why is this? Is it because I haven't adjusted the batch_size?
Fno such thing. vram has 0 impact on speed if both fitting into vram
Fit is like case of a truck
Fbatch size can certainly make impact in your case
Fbut bigger batch size = different LR
F
EYeah the A800 is ampere generation so better to compare to a 3090, while the 4090 is the newer Ada generation. Ada has more CUDA and tensor cores so more raw compute power.
FA6000 speed almost same as RTX 3090
Fas long as model fits into VRAM
D@Dr. Furkan Gözükara I have a question on regularization images, you recommend 20 to 1 repetitions always or it depends on the number of images on my dataset per subjet?
Dor also I have to set the repetition based on the number of reg images I have?
Dif I set my subjects repetitions too low the reg images will have more impact on the training than my subjets?
V@Dr. Furkan Gözükara Does SwarmUI have something like this node for flux?
Ffor every 1 step
F1 reg image
Fso 15 images 200 epochs = 3000 steps
Fyou need 3000 unique reg images
Fi dont know
V@Dr. Furkan Gözükara Do you have an epoch xyz workflow for comfyu & Flux?
Fi have for swarmui
Fnot for comfy
V@Dr. Furkan Gözükara I tried for swarm, but I get a backend error even though the comfyui server starts. Its to do with the flux model not loading.
Fi dont know works for me fine
Flol stable diffusion reddit mod team banned me because of a comment i made 5 months ago ridicilious
F
Fbefore rules and mods have been changed
NAs I know, in SD1.5, the loss value has a knowable good value when training LoRA. Does the Flux LoRA have the same good value in loss?.And then for the style portrait anime, is the loss value for each type different?
DAnyone tried this? What's the difference in settings for training b/w flux-dev & de-distill?
DDoc, the patreon page says batch size 7 is still experimental. Is that old news now? Are we getting better quality with batch size 7 than the others?
EThis is insane, I also get comments deleted when someone asks a question and I link a YouTube tutorial, yours and sometimes others because the videos "contains links to self promotion" like most of the time all the information is in the video no one is holding a gun to your head and forcing you to pay for the info.
DThey told me yesterday I should not talk about modifying comfy nodes using ChatGPT (but I think that was a member, rather than a mod) 
Floss values are really useless
Fi am about to finalize
Ffinding best checkpoints for accurate comparison
F
DAwesome!
DAn A6000 from Massed Compute can take Rank 7 training, I hope? I think they provide ample storage space.
Fsingle A6000 can do batch size 7
Fand it cant do that at lora 
Fmulti gpu kohya arrived to windows as well
Flast experiment almost completed
F