Software Engineering Courses (SECourses)•17mo ago

i found the solution myself @Dr. Furkan Gözükara

Vexmachina•8/19/24, 12:18 PM

What's the fastest way to swap between loras for use in Flux dev generations? Like Pieter Levels's photoai or danny postmaa with headshotpro, how are they swapping between the loras for all their thousands of users on the fly? I've read in the past that Pieter at least uses replicate. Are they spinning up an instance, loading something like a comfyui environment, and downloading the lora to the server every single time? There's gotta be a faster/more efficient way

VVAMP i found the solution myself @Dr. Furkan Gözükara

Furkan Gözükara SECourses•8/19/24, 12:48 PM

awesome

VVexmachina What's the fastest way to swap between loras for use in Flux dev generations? Li...

Furkan Gözükara SECourses•8/19/24, 12:52 PM

well you have to load lora to model weights and then there is unload method on diffusers. i use it on instantid latest version

Furkan Gözükara SECourses•8/19/24, 12:52 PM

you can see at source code

Furkan Gözükara SECourses•8/19/24, 12:53 PM

they probably download loras and keep in ram for a while i dont think they have active thousands of users at a time

RMK•8/19/24, 12:58 PM

@Dr. Furkan Gözükara will you be recording a tutorial for Flux kohya training?

Mattheus Chediak•8/19/24, 1:27 PM

https://www.letz.ai/blog/6_v2trainingguide

have you guys seen this, it train a image of a product to place the product in a image, i think it only can do this if the image used was generated by their model

LetzAI | Blog

LetzAI V2 training guide
| Here's what you need to consider to get the best results

Mattheus Chediak•8/19/24, 1:28 PM

im trying to find out what do they use for this

Mattheus Chediak•8/19/24, 1:33 PM

@Dr. Furkan Gözükara hey guys how can i train a model with product pictures in order to do inpainting on images with these products

i still dont understand how to do this, like placing a pop can in the hand of a model or something

Mattheus Chediak•8/19/24, 1:33 PM

Mattheus Chediak•8/19/24, 1:33 PM

or a brand t-shirt

MMattheus Chediak https://www.letz.ai/blog/6_v2trainingguide have you guys seen this, it train a ...

�

𝔸𝖨ΞӾ•8/19/24, 1:39 PM

I haven't personally tested it before but I figure the logic would be something like using a dataset of a product with the backgrounds removed, so like on a plain background and also various angles of it, and no faces whatsoever in any of the images.

Mattheus Chediak•8/19/24, 1:43 PM

i see, any more details on it? ill reall ylike to have something like this

Mattheus Chediak•8/19/24, 1:43 PM

i mean how to mix the model and th eproduct together

Mattheus Chediak•8/19/24, 1:47 PM

is it lora so i can use 10 images of the product and assk it to generate something

Sipriyani•8/19/24, 1:48 PM

I didnt exprince it either , however, what about, give a try to, rare token word (xmycompanylogo) , and train 10 images just the logo, you want to see... after training to typo , "a woman wears a green shirt that xmycompanylogo is over it. " .. after train , you may want to extract the the training as a lora.

RRMK @Dr. Furkan Gözükara will you be recording a tutorial for Flux kohya training?

Furkan Gözükara SECourses•8/19/24, 2:40 PM

yep

Furkan Gözükara SECourses•8/19/24, 2:40 PM

i am starting 8x A6000 GPU today for extensive testing

MMattheus Chediak Click to see attachment

Furkan Gözükara SECourses•8/19/24, 2:41 PM

if you can have a big dataset with proper caption training is same

MMattheus Chediak is it lora so i can use 10 images of the product and assk it to generate somethi...

Furkan Gözükara SECourses•8/19/24, 2:41 PM

if single product image yes

Furkan Gözükara SECourses•8/19/24, 3:59 PM

started flux testing

Joshcandle•8/19/24, 4:07 PM

Hello, was at vacation but back again now, do you have any training results for your Lora Training for Flux? And also how could i achive the same results with afterdetailer but for flux since it does not work in A1111

JJoshcandle Hello, was at vacation but back again now, do you have any training results for ...

Furkan Gözükara SECourses•8/19/24, 4:11 PM

swarmui segment feature

Furkan Gözükara SECourses•8/19/24, 4:12 PM

https://youtu.be/HKX8_F1Er_w

YouTubeSECourses

Zero to Hero Stable Diffusion 3 Tutorial with Amazing SwarmUI SD We...

Do not skip any part of this tutorial to master how to use Stable Diffusion 3 (SD3) with the most advanced generative AI open source APP SwarmUI. Automatic1111 SD Web UI or Fooocus are not supporting the #SD3 yet. Therefore, I am starting to make tutorials for SwarmUI as well. #StableSwarmUI is officially developed by the StabilityAI and your m...

Furkan Gözükara SECourses•8/19/24, 4:12 PM

searching flux atm

Joshcandle•8/19/24, 4:14 PM

Ah okay thx

Joshcandle•8/19/24, 4:14 PM

What u mean with seachring flux atm?

JJoshcandle What u mean with seachring flux atm?

Furkan Gözükara SECourses•8/19/24, 4:30 PM

searching parameters

ivaxsirc -4090•8/19/24, 4:38 PM

@Dr. Furkan Gözükara Forge has a new tab that's called spaces, that install separately gradio extensions, maybe you can put any oneclick installer to promote you

JJ2_Nedo if you want the best local then use Flux Dev. but not the fp8 version! plus the ...

Neuromasters•8/19/24, 5:10 PM

T5 FP16 separate download and dev version from darkwood download? I'm sorry I didn't follow sources much and I'm confused in them, there are many different models everywhere, I'll be especially grateful if you give me sources from where to download or tutor

NNeuromasters T5 FP16 separate download and dev version from darkwood download? I'm sorry I d...

J2_Nedo•8/19/24, 5:51 PM

good info for Forge here: https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/1050

GitHub

[GGUF and Flux full fp16 Model] loading T5, CLIP + new VAE UI · lll...

The old Automatic1111’s user interface of VAE selection is not powerful enough for modern models . Forge make minor modifications so that the UI is as close as possible to A1111 but also meet the d...

J2_Nedo•8/19/24, 5:52 PM

this is how it looks for me:

YellowjacketXVI•8/19/24, 6:22 PM

been training on flux and wow so great, I've been typing my captions out by hand, what tool goes well with the Clip-I?

YYellowjacketXVI been training on flux and wow so great, I've been typing my captions out by hand...

J2_Nedo•8/19/24, 6:29 PM

i use Florence 2 for this, my Flux Loras looking really good, but i am still in research mode

JJ2_Nedo i use Florence 2 for this, my Flux Loras looking really good, but i am still in ...

YellowjacketXVI•8/19/24, 6:29 PM

The man! .....not assuming anything tho..... XD

JJ2_Nedo i use Florence 2 for this, my Flux Loras looking really good, but i am still in ...

YellowjacketXVI•8/19/24, 6:31 PM

It can be used for a clip seg? How indepth of a description can you give, I have a hard time seperating one man from another with SAM

JJ2_Nedo i use Florence 2 for this, my Flux Loras looking really good, but i am still in ...

�

𝔸𝖨ΞӾ•8/19/24, 6:35 PM

I've tried Florence2 for a training and it went good, was setting up another Lora today but this time I actually read the captions first and noticed Florence got some key details very wrong, which could possibly mess up the Lora a bit so I dunno. I wonder if Llava is better or there's a free local Captioning tool that is a lot more consistent

JJ2_Nedo good info for Forge here: https://github.com/lllyasviel/stable-diffusion-webui-f...

ivaxsirc -4090•8/19/24, 6:36 PM

not works for me i see the creation image but after grey image, what's wrong?

Iivaxsirc -4090 not works for me i see the creation image but after grey image, what's wrong?

J2_Nedo•8/19/24, 6:38 PM

must be something with the sampler type, i got it too, but it works correct with the right one, try normal and others

�𝔸𝖨ΞӾ I've tried Florence2 for a training and it went good, was setting up another Lor...

J2_Nedo•8/19/24, 6:39 PM

i always read the captions and edit them manually if i see something is wrong

dsienra•8/19/24, 6:39 PM

I'm doing some testing training flux on kohya but I'm having a bleeding problem, I'm training pictures of myself as a kid and as and adult and a friend, so the captions are "myname man" "myname boy" "myfriendname man" and the concepts bleed, this problem doesn't happen training sdxl models, there is no bleeding. maybe the problem is because I'm not training the text encoder?

dsienra•8/19/24, 6:42 PM

I hade a similar problem with my sd3 testing

Ddsienra I'm doing some testing training flux on kohya but I'm having a bleeding problem,...

Furkan Gözükara SECourses•8/19/24, 6:48 PM

ye not training text encoder will cause

Furkan Gözükara SECourses•8/19/24, 6:48 PM

more bleeding

Furkan Gözükara SECourses•8/19/24, 6:48 PM

since you only train unet

Furkan Gözükara SECourses•8/19/24, 6:48 PM

clip l training not supported yet

dsienra•8/19/24, 6:49 PM

ohh thanks, so may be is that

JJ2_Nedo must be something with the sampler type, i got it too, but it works correct with...

ivaxsirc -4090•8/19/24, 6:50 PM

I've found it was the vae, are for shell not for dev

FFurkan Gözükara SECourses ye not training text encoder will cause

dsienra•8/19/24, 7:03 PM

Another question: the CLIP models, such as clip_l, are stored in separate files from the base model. If I use a concept that doesn’t exist in CLIP, such as ohwx or another non-existent word, would it still work during inference if it's not part of the model because the model was not trained? Additionally, if I train clip_l, would it be saved as a separate file? Currently, the text encoders are not in the same safetensors as the base model

Ddsienra Another question: the CLIP models, such as clip_l, are stored in separate files ...

Furkan Gözükara SECourses•8/19/24, 7:22 PM

it will work but lesser degree

i found the solution myself @Dr. Furkan Gözükara

Similar Threads