Software Engineering Courses (SECourses)•8mo ago

Ok, thank you 🙂 Only other thing I'm wondering is this: if I have a portrait image, can I just re

Ok, thank you

Only other thing I'm wondering is this: if I have a portrait image, can I just resize to 1024x1024 images and just leave the sides transparent or white/black or do I absolutely have to outpaint the sides if I want to use this image for training in my dataset?

Thanks!

EEscanor Ok, thank you 🙂 Only other thing I'm wondering is this: if I have a portrait ...

Furkan Gözükara SECourses•5/17/25, 3:40 PM

you can leave empty transparent

EscanorOP•5/17/25, 3:54 PM

oh really?? Wow.. This will make things so much easier...

I had spent to much time trying to outpaint something decent for the sides with Fooocus..

Damnit lol, glad I asked thank you very much

EEscanor oh really?? Wow.. This will make things so much easier... 🤦 I had spent to muc...

Furkan Gözükara SECourses•5/17/25, 3:57 PM

you are welcome

TheRealDude•5/17/25, 9:49 PM

TTheRealDude Hi

Furkan Gözükara SECourses•5/17/25, 10:36 PM

mlado•5/18/25, 9:48 AM

Hello everyone

I am new to the AI world, started learning around a week ago, and my goal is to create a realistic-looking Instagram influencer with the same face and body (body seems hard but at least face hah). I have a dataset of 70 images, which are all 1024x1024 and good quality I think, and I will be training a lora in Fluxgym because it seems beginner friendly.
I just wanted to ask for your suggestions and advice on a few things. What would you recommend in terms of trains per image, number of epochs, and LoRA rank for this kind of realistic, identity-focused project?

(also thank you very much for all your research and insights @Furkan Gözükara SECourses)

Mmlado Hello everyone 👋 I am new to the AI world, started learning around a week ago, ...

Timson•5/18/25, 10:07 AM

start (and end) with fine tuning, skip LORA step. It's not much harder for beginner, basically same process but drastically better in end result.
If you want to focus on the very best result for body + face train 2 separate models on 2 separate datasets - one focused on body (full, medium shots, poses) the other on face (mid, close shots), then use 2 step generation with face model over image generated on body model.
If you want easier process - balance your dataset well for close and mid shots (30% mid, 70% close)

mlado•5/18/25, 10:08 AM

Okay thank you, I have a 4070 ti super, is that sufficient for fine tuning?

mlado•5/18/25, 10:09 AM

And yeah tbh I watched doctor futkans video about fine tuning vs lora and he also said that lora is inferior

Timson•5/18/25, 10:09 AM

Idaely you would have to rent A6000, it's 0.3$ per hour and training would take 6-10 hours

mlado•5/18/25, 10:10 AM

Okay, thats through runpod right

Timson•5/18/25, 10:10 AM

no, massedcompute

mlado•5/18/25, 10:10 AM

Oh okay sorry

mlado•5/18/25, 10:11 AM

I essentially just have to follow doctor furkan’s guide right

Timson•5/18/25, 10:11 AM

that's right

mlado•5/18/25, 10:11 AM

Hah okay, it will be the more complex way then hah, but I want best results possible soo yeah

mlado•5/18/25, 10:11 AM

Thanks timson

Mmlado I essentially just have to follow doctor furkan’s guide right

Timson•5/18/25, 10:16 AM

one last advice - the end result would depend on balance and composition of training data set even more than on the process itself. Training set would represent the features you want the model to learn proportionally. The more features you want to train (body type, poses, facial expressions) the more diverce dataset you woudl need and more training steps it would take to reach optimum training point

Timson•5/18/25, 10:17 AM

Think more about balance than on amount of good quality images

mlado•5/18/25, 10:18 AM

Okay, if you have time now, would you like to hop on a call and you can have a glance at the images I have and give your opinion, I can share screen

Mmlado Okay, if you have time now, would you like to hop on a call and you can have a g...

Timson•5/18/25, 10:22 AM

sorry, that would be too much

You can order paid consultation with Dr. Furkan but I advice you to have some experience on your own first with tutorials

mlado•5/18/25, 10:23 AM

Okay, thank you regardless

Mmlado Okay thank you, I have a 4070 ti super, is that sufficient for fine tuning?

Furkan Gözükara SECourses•5/18/25, 11:14 AM

100% you can train on that gpu

Furkan Gözükara SECourses•5/18/25, 11:14 AM

if you have 64 gb ram

mlado•5/18/25, 12:37 PM

yeah thats the thing unfortunately I only have 2x16 gb 6000mhz cl30

mlado•5/18/25, 12:38 PM

if I were to upgrade my ram, can I just order another 2x16 6000mhz kit and have a total of 4x16 or will I be better off with dual channel 2x32

Mmlado if I were to upgrade my ram, can I just order another 2x16 6000mhz kit and have ...

Furkan Gözükara SECourses•5/18/25, 12:48 PM

i recommend get 2x48 if possible

Furkan Gözükara SECourses•5/18/25, 12:48 PM

4x ram usually causes too much problem

Furkan Gözükara SECourses•5/18/25, 12:48 PM

but it depends on motherboard a lot

mlado•5/18/25, 12:50 PM

oh okay, thanks for the advice, my motherboard is a new model b650 called Eagle AX by Gigabyte, I bought it for my 7800x3d so should be good enough

mlado•5/19/25, 9:26 PM

I have 70 images in my data set to create my realistic instagram ai model, how many epochs would you recomend to use when fine tuning?

mlado•5/19/25, 9:28 PM

perhaps 100?

mlado•5/19/25, 9:29 PM

I know that Dr Furkan said that best epoch is 40 when he did 256 images, and 140 when he trained 15

mlado•5/19/25, 9:29 PM

But I am still unsure hah

Mmlado I have 70 images in my data set to create my realistic instagram ai model, how m...

Furkan Gözükara SECourses•5/19/25, 11:44 PM

for 70 images go up to 150 at least

Furkan Gözükara SECourses•5/19/25, 11:44 PM

and compare

mlado•5/20/25, 9:34 AM

I have 58.6 GB of free space, will that be enough to download the SUPIR upscaler

mlado•5/20/25, 9:35 AM

im done with my training, so I wonder can I delete kohyaa to free up more space?

Coke Harbor Butcher•5/20/25, 9:41 AM

nope @mlado

mlado•5/20/25, 9:42 AM

oh ok

Mmlado I have 58.6 GB of free space, will that be enough to download the SUPIR upscaler

Furkan Gözükara SECourses•5/20/25, 10:25 AM

ye you need more. especially if that is your virtual RAM disk as well

mlado•5/20/25, 10:28 AM

Okay

mlado•5/20/25, 10:28 AM

Thank you

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭•5/20/25, 2:20 PM

Hey. My sdxl fine tune did not capture my body right. Used reg images. Now I'm thinking that since I had 25 body pics and 15 close ups of face, using ohwx woman only as tag on all pics...maybe all the close ups of face confuses or makes the fine tune forget my characters body, and is more influenced by bodies of women in reg images. Any knowledge of that?

�🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭Hey. My sdxl fine tune did not capture my body right. Used reg images. Now I'm t...

Furkan Gözükara SECourses•5/20/25, 2:21 PM

ye sdxl is weak

Furkan Gözükara SECourses•5/20/25, 2:21 PM

use flux

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭•5/21/25, 7:18 AM

https://civitai.com/models/1223425?modelVersionId=1378381 claims to have trained on flux schnell and used high resolution images, bypassing the limit of 1024px. Hmm. Can this be true, Dr. @Furkan Gözükara SECourses ? It's a sfw fine tune. No worries

TheRealDude•5/21/25, 10:10 AM

That was initially released on F1_Dev, no clue why they switched to F1_Shnell

�🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭https://civitai.com/models/1223425?modelVersionId=1378381 claims to have trained...

Furkan Gözükara SECourses•5/21/25, 10:46 PM

enabling the model to produce very fine details during tiled upscales that can hold up even in 32K upscales.

Furkan Gözükara SECourses•5/21/25, 10:46 PM

probably overfit for some classes and may work

Ok, thank you 🙂 Only other thing I'm wondering is this: if I have a portrait image, can I just re

Similar Threads