Software Engineering Courses (SECourses)•17mo ago

you told me 10,000 images is better than 1000 right

GomezBroOP•9/1/24, 9:36 PM

1000 would take like 450 hrs on my a6000 with 200 epoch

GGomezBro you told me 10,000 images is better than 1000 right

Furkan Gözükara SECourses•9/1/24, 9:36 PM

yes

Furkan Gözükara SECourses•9/1/24, 9:36 PM

if your style very consistent

Furkan Gözükara SECourses•9/1/24, 9:36 PM

even 1 epoch may be sufficient

Furkan Gözükara SECourses•9/1/24, 9:36 PM

with 10000 images

GomezBroOP•9/1/24, 9:36 PM

https://tenor.com/view/shocked-surprised-gasp-what-cat-shock-gif-635629308990545194

Tenor

GomezBroOP•9/1/24, 9:37 PM

Ohhh that explains why when I did 3000 image NFT style training on SDXL a year ago it only took 9 hrs and that lora works amazing for that art style

Furkan Gözükara SECourses•9/1/24, 9:38 PM

yep

AAfroman4peace @Dr. Furkan Gözükara I am here to say thanks for your tutorial and the kohya_ss ...

Furkan Gözükara SECourses•9/1/24, 9:38 PM

you are welcome. also this week hopefully i am fully into fine tuning FLUX

Vvirginia @Dr. Furkan Gözükara ：（ my images from swarmUI is blurry, i do not know what is ...

Furkan Gözükara SECourses•9/1/24, 9:40 PM

set CFG to 1

Furkan Gözükara SECourses•9/1/24, 9:40 PM

and do more steps

JJames K Flux can run on M3 18GB Mac ?

Furkan Gözükara SECourses•9/1/24, 9:40 PM

sadly i dont know did you try swarmui ?

Afroman4peace•9/1/24, 9:42 PM

I have created a lora that was based on the combined output of my sdxl model+ my flux model. after that I merged it with my base flux model. The result is very interesting. I use the resulting Flux model as "guidance" model for my SDXL model which I have trained over months. This allows me with just 4 flux steps and 15 SDXL steps with a simple upscale to create ultra detailed results.

Afroman4peace•9/1/24, 9:43 PM

I will use this output to finetune a proper flux model.. (instead of merging in the lora)

FFurkan Gözükara SECourses batch size 1 isnt fastest - you need to calculate effective speed

Thereiver•9/1/24, 9:44 PM

In my case it was

TThereiver In my case it was 🙂

Furkan Gözükara SECourses•9/1/24, 9:44 PM

when you increase batch size

Furkan Gözükara SECourses•9/1/24, 9:44 PM

you need to multiply step speed with it

Furkan Gözükara SECourses•9/1/24, 9:44 PM

to get practical speed

Thereiver•9/1/24, 9:45 PM

ahh makes sense

FFurkan Gözükara SECourses you are welcome. also this week hopefully i am fully into fine tuning FLUX

Afroman4peace•9/1/24, 9:48 PM

Do you think its possible to finetune a flux model with 12 GB Vram?

Thereiver•9/1/24, 9:54 PM

Pretty sure you can yea

AAfroman4peace Do you think its possible to finetune a flux model with 12 GB Vram?

Furkan Gözükara SECourses•9/1/24, 9:56 PM

we will see

Furkan Gözükara SECourses•9/1/24, 9:56 PM

there are some huge optimizations

Furkan Gözükara SECourses•9/1/24, 9:58 PM

i am working on updating flux installer

Furkan Gözükara SECourses•9/1/24, 9:58 PM

didnt notice difference with xformers

Furkan Gözükara SECourses•9/1/24, 9:58 PM

if it uses lesser vram on windows i will switch to that

Furkan Gözükara SECourses•9/1/24, 10:00 PM

i am also going to improve model downloader

Thereiver•9/1/24, 10:08 PM

My darkest dungeon lora so far with mostly your settings. Got me back in the game, thanks

Thereiver•9/1/24, 10:09 PM

Just training output images, but progress nonetheless

TThereiver My darkest dungeon lora so far with mostly your settings. Got me back in the gam...

Furkan Gözükara SECourses•9/1/24, 10:09 PM

nice

Thereiver•9/1/24, 10:10 PM

315 images, increased lr by 100x. Otherwise all settings are rank 4 i think

Thereiver•9/1/24, 10:11 PM

dim 128 and alpha 128 are working nicely

TThereiver 315 images, increased lr by 100x. Otherwise all settings are rank 4 i think

Furkan Gözükara SECourses•9/1/24, 10:11 PM

it may cook model but it depends on purpose :d

Thereiver•9/1/24, 10:12 PM

yeah probably, just experimenting to see if I can pull it off this way first

Thereiver•9/1/24, 10:14 PM

gotta relearn how to check convergence again. think i was using tensorboard

Thereiver•9/1/24, 10:15 PM

been over a year since I've done any training

Furkan Gözükara SECourses•9/1/24, 10:34 PM

new flux train models downloader

Furkan Gözükara SECourses•9/1/24, 10:34 PM

speed mind blowing

Furkan Gözükara SECourses•9/1/24, 10:34 PM

it can resume too

Furkan Gözükara SECourses•9/1/24, 11:15 PM

xformers on windows faster than sdpa

Furkan Gözükara SECourses•9/1/24, 11:15 PM

i think for low configs i will use it

Furkan Gözükara SECourses•9/1/24, 11:34 PM

xformers didnt help

Furkan Gözükara SECourses•9/1/24, 11:34 PM

testing torch 2.5 for 12gb

Konstantinas•9/1/24, 11:39 PM

I am trying to finetune Flux, but getting "no metadata" error. Do I need to somehow prepare metadata, is there any guide on that?

druhl•9/1/24, 11:50 PM

Flux is sooo good, that it is soo tough to decide which lora is best when put side to side on a generator

Ddruhl Flux is sooo good, that it is soo tough to decide which lora is best when put si...

Furkan Gözükara SECourses•9/1/24, 11:51 PM

true

KKonstantinas I am trying to finetune Flux, but getting "no metadata" error. Do I need to some...

Furkan Gözükara SECourses•9/1/24, 11:51 PM

it is my this week research hopefully

Furkan Gözükara SECourses•9/1/24, 11:51 PM

fine tuning flux

Konstantinas•9/1/24, 11:52 PM

cool

I just need to use cloth lora and character lora, and combining those degrades quality by a lot. So finetuning is only solution for me now, hence researching it... But good old Lora works perfectly when not chaining

KKonstantinas cool 🙂 I just need to use cloth lora and character lora, and combining those de...

Furkan Gözükara SECourses•9/1/24, 11:53 PM

ye true

you told me 10,000 images is better than 1000 right

Similar Threads

Similar Threads

Similar Threads