Software Engineering Courses (SECourses)•16mo ago

@Dr. Furkan Gözükara Thanks for all your hard work, following your lesson, i've trained my own model

@Dr. Furkan Gözükara Thanks for all your hard work, following your lesson, i've trained my own model based off of flux dev, and running gguf text encoder and models and made this image

2130-A20confident20zono20man20striking20a20pose20lo-FLUX1flux-hyp16-Q4_0gguf-2054978571.png

ZZono50 @Dr. Furkan Gözükara Thanks for all your hard work, following your lesson, i've ...

Furkan Gözükara SECourses•9/11/24, 1:48 AM

awesome

Furkan Gözükara SECourses•9/11/24, 1:48 AM

i plan to test loras on gguf models

Zono50OP•9/11/24, 1:59 AM

here's another one at 9:16 aspect ratio -

2155-A20confident20zono20man20striking20a20pose20lo-FLUX1flux-hyp16-Q4_0gguf-2043662712.png

Zono50OP•9/11/24, 1:59 AM

i also used your cropper and resizer

ZZono50 here's another one at 9:16 aspect ratio -

Furkan Gözükara SECourses•9/11/24, 2:15 AM

wow nice

Furkan Gözükara SECourses•9/11/24, 2:15 AM

i am making gradio for it

Furkan Gözükara SECourses•9/11/24, 2:15 AM

all in one

ZZono50 here's another one at 9:16 aspect ratio -

GomezBro•9/11/24, 2:15 AM

Nice art, big pimpin' haha

Furkan Gözükara SECourses•9/11/24, 2:56 AM

--fast brings huge speed boost on rtx 4090
from 1.6 it / second to 2.35 it / second

ZZono50 @Dr. Furkan Gözükara Thanks for all your hard work, following your lesson, i've ...

sergio_gonzo•9/11/24, 3:31 AM

you could reocmend some model thats work with loras? best regards

ZZono50 @Dr. Furkan Gözükara Thanks for all your hard work, following your lesson, i've ...

druhl•9/11/24, 4:28 AM

I still don't get the fuss about gguf models? Isn't SwarmUI capable of running the 23GB model on slower GPUs also? Is this to increase the speed of the generations? Or to be able to use more, pin-pointed comfyUI nodes?

FFurkan Gözükara SECourses --fast brings huge speed boost on rtx 4090 from 1.6 it / second to 2.35 it / sec...

protector131090•9/11/24, 5:33 AM

but its only on fp8 ? and fp8 degrades LORA likenes

Pprotector131090 but its only on fp8 ? and fp8 degrades LORA likenes

druhl•9/11/24, 5:50 AM

That might be true. I'll have to check the settings on massed compute, but perhaps I chose automatic there, and the results were kinda better than what I produce using FP8 on my home PC.

TheToday•9/11/24, 6:36 AM

What do you think about training only some layers when doing a Lora? Has anyone tried it?

PixelMonkey•9/11/24, 6:54 AM

Hey everyone, quick question, did anyone figure out a way in Flux Dev to get rid of some depth of field blur and out of focus ? I tried various approaches using keywords in the prompt but without any real success. Any advice is much appreciated.

TTheToday What do you think about training only some layers when doing a Lora? Has anyone ...

druhl•9/11/24, 6:57 AM

I hear the feature is not there on Kohya, but possible with AIToolkit, though I am yet to try that.

PPixelMonkey Hey everyone, quick question, did anyone figure out a way in Flux Dev to get rid...

protector131090•9/11/24, 6:58 AM

Boreal LORA

Ddruhl I hear the feature is not there on Kohya, but possible with AIToolkit, though I ...

TheToday•9/11/24, 6:58 AM

Yes, is in AIToolKit

Pprotector131090 Boreal LORA

PixelMonkey•9/11/24, 6:59 AM

Thx, will test that, where can I find that one.. Civit? EDIT: found it i think . Much appreciated

PixelMonkey•9/11/24, 7:06 AM

Other quick question, is there a way to test mutliple loras in comparison in Stable Diffusion Forge webUi via script , for example X/Y/Z matrix or similar?

TheToday•9/11/24, 7:08 AM

Yes you can do with xyz and select prompt S/R

PPixelMonkey Other quick question, is there a way to test mutliple loras in comparison in Sta...

protector131090•9/11/24, 7:09 AM

in forge it works same as in A1111 and in comfy its also posible but its hard to make it work

Pprotector131090 in forge it works same as in A1111 and in comfy its also posible but its hard to...

PixelMonkey•9/11/24, 7:10 AM

Is there anything to watch out for if you say its hard to make it work ?

PixelMonkey•9/11/24, 7:39 AM

What are the best settings on a rtx 3090? feels some times very slow and other times it runs smooth ... bit weird

PPixelMonkey Is there anything to watch out for if you say its hard to make it work ?

protector131090•9/11/24, 7:58 AM

im just not good at comfy. it could be easy. but i couldnt figure it out...

PixelMonkey•9/11/24, 8:00 AM

Same here.. either render gets stuck

or doesnt work.. in forge.. not good in comfy better said comfortable to try that, and setup a workflow.. just want to add more realism and compare the loras for best result.. maybe i check swarm UI ?