Software Engineering Courses (SECourses)•16mo ago

Hi @Dr. Furkan Gözükara for training person you said that no caption is better than with caption, ho

Hi @Dr. Furkan Gözükara for training person you said that no caption is better than with caption, however when I read your 3d lora style training section you also said if the dataset is inconsistent, caption is better. So does it apply to training person too

wardensc2OP•9/13/24, 5:14 AM

My dataset is something like this

wardensc2OP•9/13/24, 5:15 AM

multiple size with 1024x1024, 512x768 or 768x512

wardensc2OP•9/13/24, 5:15 AM

Does it count as inconsistent dataset

wardensc2OP•9/13/24, 5:15 AM

Do I need to train with caption

FFurkan Gözükara SECourses it is format of swarmui

PixelMonkey•9/13/24, 7:15 AM

havent tried Swarm UI .. versy used to SD Forge . But maybw worth looking into ?

МИШАНЯ•9/13/24, 7:46 AM

flux allows to create in images inscriptions in English, is there a tool to add inscriptions in Russian?

AiMan1993•9/13/24, 9:45 AM

Hi Everyone

AiMan1993•9/13/24, 9:46 AM

who got experience with using AI Toolkit for fine tuning flux?

AiMan1993•9/13/24, 9:46 AM

I need some guidance please!

Wwardensc2 Hi @Dr. Furkan Gözükara for training person you said that no caption is better t...

Furkan Gözükara SECourses•9/13/24, 10:35 AM

good question. it is a same person so may not apply

Furkan Gözükara SECourses•9/13/24, 10:35 AM

unless person is very different in photos

PPixelMonkey havent tried Swarm UI .. versy used to SD Forge . But maybw worth looking into ?

Furkan Gözükara SECourses•9/13/24, 10:35 AM

yes swarm is really good

ММИШАНЯ flux allows to create in images inscriptions in English, is there a tool to add ...

Furkan Gözükara SECourses•9/13/24, 10:35 AM

not that i know but fine tuning may add that capability

AAiMan1993 Hi Everyone

Furkan Gözükara SECourses•9/13/24, 10:36 AM

why dont you use kohya?

FFurkan Gözükara SECourses not that i know but fine tuning may add that capability

МИШАНЯ•9/13/24, 10:36 AM

What do you mean? I'm not very good at this...

ММИШАНЯ What do you mean? I'm not very good at this...

Furkan Gözükara SECourses•9/13/24, 10:37 AM

i mean model need to be trained on such dataset

�

𝗡𝗶𝘁𝗶𝗻 𝗣𝗮𝗿𝗲𝘃𝗮•9/13/24, 10:56 AM

@Dr. Furkan Gözükara Can I set the weight of lora used during generating samples while it's training or anyway to know what weight it used to generate those samples?

FFurkan Gözükara SECourses why dont you use kohya?

AiMan1993•9/13/24, 11:04 AM

Sure I can give try

AiMan1993•9/13/24, 11:04 AM

I want to replicate a meme actually - naming Kwen

AiMan1993•9/13/24, 11:05 AM

For e.g. this character

AiMan1993•9/13/24, 11:05 AM

AiMan1993•9/13/24, 11:05 AM

I took this from net but lets assume this is not there in the net

AiMan1993•9/13/24, 11:05 AM

and the model havent seen the character at all

AiMan1993•9/13/24, 11:06 AM

so what configuarations will give me the best results

AiMan1993•9/13/24, 11:06 AM

How many pictures I will need

�𝗡𝗶𝘁𝗶𝗻 𝗣𝗮𝗿𝗲𝘃𝗮 @Dr. Furkan Gözükara Can I set the weight of lora used during generating samples...

Furkan Gözükara SECourses•9/13/24, 11:07 AM

i dont use samples usually not working

Furkan Gözükara SECourses•9/13/24, 11:07 AM

or breaking training :d

AiMan1993•9/13/24, 11:07 AM

So model has no information about Kwen it doesnt know anything but I want to generate such that it hangs out with Mickey Mouse

AAiMan1993 So model has no information about Kwen it doesnt know anything but I want to gen...

Furkan Gözükara SECourses•9/13/24, 11:07 AM

with dataset you can make

AiMan1993•9/13/24, 11:08 AM

So how many pictures will I need? LoRA ranking? Captioning of image? There is florence 2 built in with ai toolkit but isnt great at captioning

AAiMan1993 So how many pictures will I need? LoRA ranking? Captioning of image? There is fl...

Furkan Gözükara SECourses•9/13/24, 11:20 AM

more better

Furkan Gözükara SECourses•9/13/24, 11:20 AM

we have full configs on patreon use according to your vram

Furkan Gözükara SECourses•9/13/24, 11:20 AM

i dont suggeest captions

AiMan1993•9/13/24, 12:08 PM

yes I have subscribed to your patreon

AiMan1993•9/13/24, 12:34 PM

Will I be able to use it with ComfyUI?

AiMan1993•9/13/24, 12:34 PM

after fine tuning with Kohya?

AAiMan1993 after fine tuning with Kohya?

Furkan Gözükara SECourses•9/13/24, 12:46 PM

of course

Furkan Gözükara SECourses•9/13/24, 12:46 PM

i use with swarmui it uses comfyui as back end

AiMan1993•9/13/24, 12:47 PM

Nice!

�

𝗡𝗶𝘁𝗶𝗻 𝗣𝗮𝗿𝗲𝘃𝗮•9/13/24, 1:05 PM

@Dr. Furkan Gözükara What to do about bleeding? I have tried lower network and lower learning rates but can't stop bleeding

TTOPUI If my current image size is 16:9 and the resolution is 1024X576, is this how the...

Kallamamran•9/13/24, 1:18 PM

No. You only specify one value for vertical AND horizontal resolution. The training is on square images. You should specify 1024 in your case. The images will be bucketed if they ar not square or if they are a different (higher) resolution than what's being trained.

Kallamamran•9/13/24, 1:19 PM

@Dr. Furkan Gözükara https://huggingface.co/blog/MonsterMMORPG/ultimate-flux-lora-training-tutorial <- Is 404

FFurkan Gözükara SECourses yes dataset is king

sergio_gonzo•9/13/24, 1:30 PM

Dr Furkan there is new settinsg that improve the velocity of trianing for a 4070 ti super 16 vram actualli is 10 its i don rmeember the nomenclaure in rank 5

�𝗡𝗶𝘁𝗶𝗻 𝗣𝗮𝗿𝗲𝘃𝗮 @Dr. Furkan Gözükara What to do about bleeding? I have tried lower network and l...

Furkan Gözükara SECourses•9/13/24, 1:44 PM

i couıldnt stop yet either