Software Engineering Courses (SECourses)•8mo ago

what ratio you recommend for full or half body training or is it best to crop them manually to 1024

what ratio you recommend for full or half body training or is it best to crop them manually to 1024 x 1024 with the most of body and face what is possible

SakkimuchOP•5/6/25, 3:22 PM

if it gets best quality with 1024px i cropped it no problem

SSakkimuch what ratio you recommend for full or half body training or is it best to crop th...

Furkan Gözükara SECourses•5/6/25, 3:24 PM

i trained 1024x1024 it works perfect

Furkan Gözükara SECourses•5/6/25, 3:24 PM

as long as your dataset good it will work

SakkimuchOP•5/6/25, 3:28 PM

ok last question: would you recommend captions. I think I read in one of your posts that it doesn't help with LORAs, is that still the case?

SakkimuchOP•5/6/25, 3:28 PM

specific on person loras

SSakkimuch ok last question: would you recommend captions. I think I read in one of your po...

Furkan Gözükara SECourses•5/6/25, 3:33 PM

nope

Furkan Gözükara SECourses•5/6/25, 3:33 PM

just ohwx man or ohwx woman

SakkimuchOP•5/6/25, 3:37 PM

OK thank you so much i appreciate your work and this much Research

SSakkimuch OK thank you so much i appreciate your work and this much Research 😀

Furkan Gözükara SECourses•5/6/25, 3:55 PM

ty too

skyrrr•5/7/25, 9:16 AM

Hey guys, can we use a fine-tune flux checkpoint for impainting ? Or not recommended ?

Sskyrrr Hey guys, can we use a fine-tune flux checkpoint for impainting ? Or not recomme...

Furkan Gözükara SECourses•5/7/25, 10:10 AM

yes

Furkan Gözükara SECourses•5/7/25, 10:10 AM

you can use

N7-Blackbird•5/7/25, 10:24 AM

@Furkan Gözükara SECourses is there a tutorial or explaination on how to determine if the model is overfit or not? I know it was mentioned somewhere, but I cant find a specific clear instructions. Thank you!

NN7-Blackbird <@205854764540362752> is there a tutorial or explaination on how to determine if...

Furkan Gözükara SECourses•5/7/25, 12:04 PM

only way is generating samples

Furkan Gözükara SECourses•5/7/25, 12:04 PM

sadly loss rate is meaningless

TrumpYouAll•5/7/25, 1:48 PM

@Furkan Gözükara SECourses Will you make a video to tell us how to train HiDream by using Dreambooth in MassedCompute Cloud?

TTrumpYouAll <@205854764540362752> Will you make a video to tell us how to train HiDream by u...

Furkan Gözükara SECourses•5/7/25, 1:51 PM

yes i plan to. the things that comes never ending :d

FFurkan Gözükara SECourses yes i plan to. the things that comes never ending :d

TrumpYouAll•5/7/25, 1:52 PM

Thanks very much, I'd like to use HiDream but without Dreambooth, not much use for me for now.

I know you are gonna make it ! Thanks @Furkan Gözükara SECourses

skyrrr•5/7/25, 1:55 PM

Do you think it worth it compared to flux dev finetune ?

Sskyrrr Do you think it worth it compared to flux dev finetune ?

TrumpYouAll•5/7/25, 2:12 PM

Flux has too many internal censorships

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭•5/7/25, 2:58 PM

Has anyone fine tuned on pixelwave checkpoint?

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭•5/7/25, 2:58 PM

Or made lora

�🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭Has anyone fine tuned on pixelwave checkpoint?

Furkan Gözükara SECourses•5/7/25, 5:10 PM

i did

Furkan Gözükara SECourses•5/7/25, 5:10 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…

Sskyrrr Do you think it worth it compared to flux dev finetune ?

Furkan Gözükara SECourses•5/7/25, 5:11 PM

what i care is multi concept

Furkan Gözükara SECourses•5/7/25, 5:11 PM

if works or not

Furkan Gözükara SECourses•5/7/25, 5:11 PM

otherwise for single concept flux pretty much amazingly working

Escanor•5/8/25, 9:38 PM

Hello, I want to train sdxl koyha_ss on massed compute with A100 graphic card, I use the Tier1_48_GB_Faster_v2 config but I don't know what epoch I should set. I only have 8 images and I use all 5000 reg images. From what I understand, I should aim for 3000-5000 total steps. So I'm currently trying with the default settings of the configuration but setting 400 repeats so I have about 3200 steps...

Is 400 repeats with 1 epoch and 8 pictures good?
I set a batch size of 4 since the A100 gpu has 80gb of memory. Is this right?
Could I optimize the config better to be faster or more efficient for A100 gpu?

Thank you very much for all the tutorials and the help

Escanor•5/8/25, 9:43 PM

Also, on massed compute, I see Could not find TensorRTCould not find TensorRT when starting training with kohya_ss, is this normal? Does it matter to install TensorRT for training only?

Ah and if it's only 8 portrait images, do I want captions? From what I understand no

Thanks

EEscanor Hello, I want to train sdxl koyha_ss on massed compute with A100 graphic card, I...

Furkan Gözükara SECourses•5/8/25, 10:33 PM

welcome

EEscanor Also, on massed compute, I see `Could not find TensorRT` when starting training ...

Furkan Gözükara SECourses•5/8/25, 10:33 PM

not important

Furkan Gözükara SECourses•5/8/25, 10:33 PM

8 images really low

Furkan Gözükara SECourses•5/8/25, 10:33 PM

so go up to 300 epochs

Furkan Gözükara SECourses•5/8/25, 10:33 PM

since it is sdxl

Furkan Gözükara SECourses•5/8/25, 10:33 PM

make sure to use reg images

Furkan Gözükara SECourses•5/8/25, 10:34 PM

for kohya

Furkan Gözükara SECourses•5/8/25, 10:34 PM

you need to test repeating to be able to use class images

Furkan Gözükara SECourses•5/8/25, 10:34 PM

400 repeat 1 epoch sounding good

Furkan Gözükara SECourses•5/8/25, 10:34 PM

make sure to save once every n steps

FFurkan Gözükara SECourses make sure to save once every n steps

Escanor•5/8/25, 10:51 PM

Ok, and is the tier1 48gb config optimal for A100 graphic card with 80gb or I should change something to make it better and use more available memory? Batch size around 4-7? Do you recommend onetrainer finetune over kohya_ss dreambooth?

Thank you for the quick answer

EEscanor Ok, and is the tier1 48gb config optimal for A100 graphic card with 80gb or I sh...

Furkan Gözükara SECourses•5/8/25, 10:53 PM

i dont recommend change anything it is best quality

Escanor•5/8/25, 11:01 PM

I'll try to give results here if I find the best sweet spot for my number of training images!

I do think it could be useful to have somer kind of pinned message with recommended setups of epoch #, repeat # and such specific to the # of images, or at least a ballpark close to it

Like 8 images: 400 repeats, 1 epoch

EEscanor Ok 🙂 I'll try to give results here if I find the best sweet spot for my number...

Furkan Gözükara SECourses•5/8/25, 11:13 PM

sadly there isnt much but i usually recommend 200 epoch

Furkan Gözükara SECourses•5/8/25, 11:13 PM

however your train image count too low

Furkan Gözükara SECourses•5/8/25, 11:14 PM

more is better

Escanor•5/8/25, 11:39 PM

Yeah I only have 8 images, I'm hoping I can finetune a model and generate decent enough images to be able to train with more images... Not sure how that will go lol

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭•5/9/25, 9:08 AM

hi, for quick trainings I often use civitai. I remember you talked about rank of lora in kohya, here it is both rank and alpha, and not sure about these. The default seems to be the same for both flux and sdxl trainings: Default is also not adafactor, but the Adam...any recommendations here? realistic person training

EEscanor Yeah I only have 8 images, I'm hoping I can finetune a model and generate decent...

Furkan Gözükara SECourses•5/9/25, 1:52 PM

you can also do this

Furkan Gözükara SECourses•5/9/25, 1:52 PM

1st fine tune generate pick good ones

Furkan Gözükara SECourses•5/9/25, 1:52 PM

and fine tune again

what ratio you recommend for full or half body training or is it best to crop them manually to 1024

Similar Threads