Software Engineering Courses (SECourses)•15mo ago

I know you've provided scripts/tools to id faces and crop etc, is there something similar that will

I know you've provided scripts/tools to id faces and crop etc, is there something similar that will just expand the canvas of an image to make it square?

OOinkSauce I know you've provided scripts/tools to id faces and crop etc, is there somethin...

Furkan Gözükara SECourses•10/11/24, 1:28 PM

it does that

Furkan Gözükara SECourses•10/11/24, 1:28 PM

if image resolution is big

Furkan Gözükara SECourses•10/11/24, 1:28 PM

and if can do without cropping subject

Furkan Gözükara SECourses•10/11/24, 1:28 PM

so that is why crop first

Furkan Gözükara SECourses•10/11/24, 1:28 PM

resize second

OinkSauceOP•10/11/24, 1:29 PM

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?

OOinkSauce so if there's a series of portrait style images, it will expand the canvas to ma...

Furkan Gözükara SECourses•10/11/24, 1:43 PM

if that image

Furkan Gözükara SECourses•10/11/24, 1:44 PM

has person in distant

Furkan Gözükara SECourses•10/11/24, 1:44 PM

so that canvas can be expanded

Furkan Gözükara SECourses•10/11/24, 1:44 PM

if not it will crop

Furkan Gözükara SECourses•10/11/24, 1:44 PM

i think i added expand with white background too

Noel Camus•10/11/24, 1:49 PM

FFurkan Gözükara SECourses your speed totally good

Noel Camus•10/11/24, 1:57 PM

Shouldn'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?

NNoel Camus Shouldn't larger VRAM bring more speed? The A800 (using Rank 1) has more than 3 ...

Furkan Gözükara SECourses•10/11/24, 1:58 PM

no such thing. vram has 0 impact on speed if both fitting into vram

Furkan Gözükara SECourses•10/11/24, 1:58 PM

it is like case of a truck

Furkan Gözükara SECourses•10/11/24, 1:58 PM

batch size can certainly make impact in your case

Furkan Gözükara SECourses•10/11/24, 1:58 PM

but bigger batch size = different LR

Furkan Gözükara SECourses•10/11/24, 2:19 PM

invoke ai broken : https://github.com/invoke-ai/InvokeAI/issues/7091

NNoel Camus Shouldn't larger VRAM bring more speed? The A800 (using Rank 1) has more than 3 ...

Error-404-unknown•10/11/24, 2:20 PM

Yeah 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.

EError-404-unknown Yeah the A800 is ampere generation so better to compare to a 3090, while the 409...

Furkan Gözükara SECourses•10/11/24, 2:21 PM

A6000 speed almost same as RTX 3090

Furkan Gözükara SECourses•10/11/24, 2:21 PM

as long as model fits into VRAM

dsienra•10/11/24, 4:43 PM

@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?

dsienra•10/11/24, 4:54 PM

or also I have to set the repetition based on the number of reg images I have?

dsienra•10/11/24, 4:57 PM

if I set my subjects repetitions too low the reg images will have more impact on the training than my subjets?

Vigilence•10/11/24, 8:22 PM

@Dr. Furkan Gözükara Does SwarmUI have something like this node for flux?

Ddsienra @Dr. Furkan Gözükara I have a question on regularization images, you recommend 2...

Furkan Gözükara SECourses•10/11/24, 8:24 PM

for every 1 step

Furkan Gözükara SECourses•10/11/24, 8:25 PM

1 reg image

Furkan Gözükara SECourses•10/11/24, 8:25 PM

so 15 images 200 epochs = 3000 steps

Furkan Gözükara SECourses•10/11/24, 8:25 PM

you need 3000 unique reg images

VVigilence @Dr. Furkan Gözükara Does SwarmUI have something like this node for flux?

Furkan Gözükara SECourses•10/11/24, 8:25 PM

i dont know

Vigilence•10/11/24, 8:47 PM

@Dr. Furkan Gözükara Do you have an epoch xyz workflow for comfyu & Flux?

VVigilence @Dr. Furkan Gözükara Do you have an epoch xyz workflow for comfyu & Flux?

Furkan Gözükara SECourses•10/11/24, 9:07 PM

i have for swarmui

Furkan Gözükara SECourses•10/11/24, 9:07 PM

not for comfy

Vigilence•10/11/24, 9:28 PM

@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.

VVigilence @Dr. Furkan Gözükara I tried for swarm, but I get a backend error even though th...

Furkan Gözükara SECourses•10/11/24, 10:17 PM

i dont know works for me fine

Furkan Gözükara SECourses•10/11/24, 10:49 PM

lol stable diffusion reddit mod team banned me because of a comment i made 5 months ago ridicilious

Furkan Gözükara SECourses•10/11/24, 10:49 PM

Furkan Gözükara SECourses•10/11/24, 10:49 PM

before rules and mods have been changed

Noel Camus•10/12/24, 4:32 AM

As 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?

Ddsienra @Dr. Furkan Gözükara do you know that parameter function? I'm training flux-dev-...

druhl•10/12/24, 4:45 AM

Anyone tried this? What's the difference in settings for training b/w flux-dev & de-distill?

FFurkan Gözükara SECourses also i sugest fine tuning batch size 7

druhl•10/12/24, 5:26 AM

Doc, 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?

FFurkan Gözükara SECourses lol stable diffusion reddit mod team banned me because of a comment i made 5 mon...

Error-404-unknown•10/12/24, 6:38 AM

This 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.

EError-404-unknown This is insane, I also get comments deleted when someone asks a question and I l...

druhl•10/12/24, 8:00 AM

They told me yesterday I should not talk about modifying comfy nodes using ChatGPT (but I think that was a member, rather than a mod)