Software Engineering Courses (SECourses)•10mo ago

whenever my model puts other people in the scene with faces visible one of them always also has my m

whenever my model puts other people in the scene with faces visible one of them always also has my model's face even if their body is the opposite sex

PPlutoISaPlanet whenever my model puts other people in the scene with faces visible one of them ...

Furkan Gözükara SECourses•3/25/25, 4:44 PM

yes true :d

Furkan Gözükara SECourses•3/25/25, 4:44 PM

this is flux bleed problem

Furkan Gözükara SECourses•3/25/25, 4:44 PM

very famous

Furkan Gözükara SECourses•3/25/25, 4:44 PM

https://medium.com/@furkangozukara/thoroughly-experimented-with-fine-tuning-dreambooth-training-of-flux-dev-de-distill-pixelwave-13fab8c625bf

Medium

Thoroughly experimented with Fine-Tuning / DreamBooth training of F...

Thoroughly experimented with Fine-Tuning / DreamBooth training of Flux-dev-de-distill, PixelWave v03, Verus Vision and base FLUX Dev model…

PlutoISaPlanetOP•3/25/25, 6:05 PM

too funny

Kaiden Quinart•3/26/25, 5:59 AM

Anyone have any ideas about training a Lora on a place. I mean teaching the model about a place to generate as background?

KKaiden Quinart Anyone have any ideas about training a Lora on a place. I mean teaching the mode...

Furkan Gözükara SECourses•3/26/25, 12:29 PM

the logic sounds same

Furkan Gözükara SECourses•3/26/25, 12:30 PM

just ohwx and the place class

Furkan Gözükara SECourses•3/26/25, 12:30 PM

then during inference prompt in details that would describe that place

Kaiden Quinart•3/26/25, 2:28 PM

with 0.0003 LR adamw_8bit do you think 200 epoch for 20 image will make it over trainned?

KKaiden Quinart with 0.0003 LR adamw_8bit do you think 200 epoch for 20 image will make it over ...

Furkan Gözükara SECourses•3/26/25, 2:28 PM

i cant comment based on other configs. but my config wont

Kaiden Quinart•3/26/25, 2:31 PM

I just found that if you put "from text file per sample" , and dont provide any caption, it still does the lora training. what will it actually do?

Kaiden Quinart•3/26/25, 2:31 PM

I tried that and it was working without any trigger word. like textual embedding

KKaiden Quinart I just found that if you put "from text file per sample" , and dont provide any ...

Furkan Gözükara SECourses•3/26/25, 2:42 PM

it will use internal caption it has

Furkan Gözükara SECourses•3/26/25, 2:42 PM

so will work still true

Deleted User•3/29/25, 7:41 AM

I've used your Flux dreambooth training presets and the results are amazing. Would training at 1280 x 1280 increase the quality even further or is that useless to try?

Deleted User•3/29/25, 9:36 AM

well I'm trying it anyway lol but it's going to take awhile

DDeleted User I've used your Flux dreambooth training presets and the results are amazing. Wo...

Furkan Gözükara SECourses•3/29/25, 12:13 PM

sadly not improving

DDeleted User well I'm trying it anyway lol but it's going to take awhile

Furkan Gözükara SECourses•3/29/25, 12:13 PM

let me know results. but i didnt get better

nicearoni•3/29/25, 11:00 PM

I usually do LoRAs over full dreambooth trainings. did one LoRA at 1536x1536. on my 3090 i did 150 epochs on 20 images and it took 22 hours

same dataset and settings at 1024x1024 had only taken 7 hours. (i think the configs and kohya itself have both been optimized a lot since then) i do feel like the 1536 LoRA is better, more detailed. but after that i stuck with 1024 - didn't feel like the improvement (it's usually subjective by then anyway with character LoRAs from only epoch 100-125 turning out awesome as well) were worth triple the training time.

Nnicearoni I usually do LoRAs over full dreambooth trainings. did one LoRA at 1536x1536. on...

Furkan Gözükara SECourses•3/29/25, 11:01 PM

yes doesnt help much. but you can try tiled training

Furkan Gözükara SECourses•3/29/25, 11:01 PM

may work for objects

Deleted User•3/30/25, 1:22 AM

I trained 59 images at 1024 x 1024 190 epochs (of a complex artstyle) and it works great even at high resolutions like 2000x2500 or something without upscaling. So I'm really curious to see if training at 1280x1280 would offer any improvements at high resolutions. I'll let you know whenever it finishes.

DDeleted User I trained 59 images at 1024 x 1024 190 epochs (of a complex artstyle) and it wor...

Furkan Gözükara SECourses•3/30/25, 11:04 AM

if you test let me know please ty

Deleted User•3/30/25, 6:05 PM

I'll be able to check in a few hours

Deleted User•3/31/25, 2:25 AM

Some things look better, some things look worse. I think it's not really worth it...
It seems to add more detail to small text or backgrounds, but the characters look worse

DDeleted User Some things look better, some things look worse. I think it's not really worth i...

Furkan Gözükara SECourses•3/31/25, 8:59 AM

thanks a lot for info

cyberbol•3/31/25, 3:43 PM

I rented massed compute first time after few month off since i have rtx 4090... I though generating wan2.1 will be much faster....lol.. definitely not on L40S lol slow.. Maybe H100, but to expensive..

Ccyberbol I rented massed compute first time after few month off since i have rtx 4090... ...

Furkan Gözükara SECourses•3/31/25, 9:01 PM

true. 4090 is still a beast

Furkan Gözükara SECourses•3/31/25, 9:01 PM

and china selling 48gb and 96 gb . 48 gb is literally l40s

Kevin•4/1/25, 4:41 AM

man I'm trying to do dreamboothsdxl training to the likeness of an individual and I've tried so many times and the sample output images continue sucking. I subscribed to the Patreon, can anyone point me towards the most useful pages re: optimal parameters and so on?

Kevin•4/1/25, 5:08 AM

If your training data set is 59 images and each is being repeated 10 times, how many regularization images is ideal? Also, is repeating 10 times the best number of repeats?

KKevin If your training data set is 59 images and each is being repeated 10 times, how ...

Furkan Gözükara SECourses•4/1/25, 10:10 PM

hello sorry for late reply

Furkan Gözükara SECourses•4/1/25, 10:10 PM

i recommend you to train flux and we dont use reg images there

Furkan Gözükara SECourses•4/1/25, 10:10 PM

here best tutorial

Furkan Gözükara SECourses•4/1/25, 10:11 PM

https://youtu.be/FvpWy1x5etM

YouTubeSECourses

FLUX Full Fine-Tuning / DreamBooth Training Master Tutorial for Win...

If you want to train FLUX with maximum possible quality, this is the tutorial looking for. In this comprehensive tutorial, you will learn how to install Kohya GUI and use it to fully Fine-Tune / DreamBooth FLUX model. After that how to use SwarmUI to compare generated checkpoints / models and find the very best one to generate most amazing image...

Kevin•4/1/25, 10:12 PM

Thanks apprecaiate it, I have since switrched to flux and am working on that currently.

Kevin•4/1/25, 10:12 PM

Would you say FLUX is the current SOTA for photoealistic images trained to a particular individual?

KKevin Would you say FLUX is the current SOTA for photoealistic images trained to a par...

Furkan Gözükara SECourses•4/1/25, 10:14 PM

100%

Kevin•4/1/25, 10:15 PM

I bought the personal consultation thing at Patreon, how do we set that up?

KKevin I bought the personal consultation thing at Patreon, how do we set that up?

Furkan Gözükara SECourses•4/1/25, 10:17 PM

what is your username?

Furkan Gözükara SECourses•4/1/25, 10:18 PM

you can private message me here

Furkan Gözükara SECourses•4/1/25, 10:18 PM

ok i see it

Furkan Gözükara SECourses•4/1/25, 10:18 PM

message me here so we can setup. we can do now too

saberu0126•4/3/25, 4:19 AM

I did too- Gold membership

Ssaberu0126 I did too- Gold membership

Furkan Gözükara SECourses•4/3/25, 11:46 AM

hello please message me details . did you subscribe gold? private message me

moomoowowcow•4/3/25, 3:59 PM

Are there any special instructions or 'Additional parameters' to enter into the Kohya_ss GUI when trying to "Resume from saved training state (path to "last-state" state folder)?"

I have entered the path/workspace/stable-diffusion-webui-forge/models/Stable-diffusion/fruit_realismByStableYogi_v5XLFP16/model/ohwx-woman_Fruit_Full-finetune_adafactor_realismIllustriousBy_v35FP16_292-Img-200-Instances-step00032000-state

I have verified the contents within that folder include all of the expected, model_1, model, optimizer, random_states, scheduler.

The training starts, however, it starts at step 01, versus step 32001 as I intend to continue traning.

Can anyone offer insight or help here? Thank you in advance!

Q•4/3/25, 10:27 PM

does kohya work with rtx 5090?

QQ does kohya work with rtx 5090? 🤔

Furkan Gözükara SECourses•4/3/25, 11:00 PM

yes

Furkan Gözükara SECourses•4/3/25, 11:00 PM

we have updated our installers just use it

whenever my model puts other people in the scene with faces visible one of them always also has my m

Similar Threads