Software Engineering Courses (SECourses)•14mo ago

Has anyone compared dreambooth 1:1 aspect training vs bucket training yet for Flux? Please let me kn

Has anyone compared dreambooth 1:1 aspect training vs bucket training yet for Flux? Please let me know your thoughts. ( alrdy know your opinion Dr Furkan

)

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭•11/23/24, 8:09 AM

guys, say I have 1000 great photos of 1 character, is it always better to use more photos, or should I stick to 256 or fewer?

�🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭guys, say I have 1000 great photos of 1 character, is it always better to use mo...

XCogni•11/23/24, 8:14 AM

for a person max 75

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭•11/23/24, 8:25 AM

oh really? how do you know

XCogni•11/23/24, 8:30 AM

trained 100s of loras

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭•11/23/24, 8:35 AM

ok, but I'm fine-tuning checkpoint, and the same applies there?

XCogni•11/23/24, 8:39 AM

yes since its only one character

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭•11/23/24, 8:54 AM

thank you

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭•11/23/24, 9:32 AM

Can I dm you?

XCogni•11/23/24, 9:50 AM

sure you can dm me @🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭

Felipe Machado•11/23/24, 1:56 PM

Doc , how many images in your data set did you use to train this beauty ?

Felipe Machado•11/23/24, 1:56 PM

FFelipe Machado Doc , how many images in your data set did you use to train this beauty ?

Furkan Gözükara SECourses•11/24/24, 12:11 AM

i plan to publish a good tutorial for this but for higher tier or sell based with dataset

Furkan Gözükara SECourses•11/24/24, 12:11 AM

what matter is dataset and how you caption

Furkan Gözükara SECourses•11/24/24, 12:11 AM

but our configs are 100% same

Nnick Has anyone compared dreambooth 1:1 aspect training vs bucket training yet for Fl...

wardensc2•11/24/24, 3:25 AM

I trained bucket from the beginning so far so good but I haven't trained 1:1 so dont know about compare result yet

ArtChicken•11/24/24, 11:50 AM

I repost also here... maybe it's the correct section:
Hi Guys... it's black-friday/cybermonday time... do you have any advice on what to buy to make flux/sdxl/pdxl models locally? I mean: what minimum requirements for a graphic card that can handle those kind of jobs... something cheap, as I do training only to understand, not for selling my works around, so I can bear also longest trainings... woud appreciate model cards or laptops adequately equipped advices, thanks to all!

dxqb•11/24/24, 12:48 PM

I went for cheapest CPU cheapest PC with 4070 Ti Super 16 GB and it's holding up. I'd go for 64 GB RAM now though because the latest offloading techniques by kohya and OneTrainer need lots of RAM.
everything can change though by tomorrow when a new model comes out. but for the current models this is a very cost-effective setup

Ddxqb I went for cheapest CPU cheapest PC with 4070 Ti Super 16 GB and it's holding up...

Furkan Gözükara SECourses•11/24/24, 2:06 PM

so 100% 64gb definitely needed

FFurkan Gözükara SECourses so 100% 64gb definitely needed

dxqb•11/24/24, 2:10 PM

question? or agreement?

dxqb•11/24/24, 2:13 PM

btw @Furkan Gözükara SECourses since you've complained about the OneTrainer UI on massed compute again: the cloud extension is very mature by now. a few people use it. you can run OneTrainer locally but train remotely. you might want to try it. it works best on runpod, but also works with any other cloud that gives SSH access

dxqb•11/24/24, 2:18 PM

also for multiple parallel runs

Ddxqb also for multiple parallel runs

Furkan Gözükara SECourses•11/24/24, 2:50 PM

nice . i didnt have chance to test it yet

FFurkan Gözükara SECourses nice . i didnt have chance to test it yet

dxqb•11/24/24, 2:52 PM

join us here for questions & feedback when you have the time https://discord.com/channels/1102003518203756564/1310219257472221255

Konstantinas•11/24/24, 3:49 PM

I want to finetune SD 3.5 to teach it 10 different characters, what is the best tool for this atm - kohya, simple tuner, ai toolkit?

KKonstantinas I want to finetune SD 3.5 to teach it 10 different characters, what is the best ...

Furkan Gözükara SECourses•11/24/24, 3:58 PM

kohya has support but i never used yet for sd 3.5

Furkan Gözükara SECourses•11/24/24, 3:58 PM

on my next to do list hopefully

Furkan Gözükara SECourses•11/24/24, 3:58 PM

currently i am trying to teach 2 character at the same time

Furkan Gözükara SECourses•11/24/24, 3:58 PM

without bleed to flux

Konstantinas•11/24/24, 3:58 PM

got it, thanks, I'll google for tutorials online to have some starting point then and will see how it goes

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭•11/24/24, 10:30 PM

if I merge my fine-tune flux dev checkpoint with another checkpoint, will my trigger words loose their effect? ohwx woman

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭•11/24/24, 10:32 PM

during the grid test in swarm, is it possible to get full quality images? I'm getting jpg ones, I even changed to png in the settings. It's like during grid, the pic quality is less good.

�🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭if I merge my fine-tune flux dev checkpoint with another checkpoint, will my tri...

Furkan Gözükara SECourses•11/24/24, 10:43 PM

yes

Furkan Gözükara SECourses•11/24/24, 10:43 PM

it will lose impact

�

🍭🎀 𝒜𝓋𝒶 𝐹𝓇𝒾𝑔𝑔 🎀🍭•11/24/24, 10:45 PM

Thank you, Dr. I thought so. Way to go is using my checkpoint, converting the other one to lora then generate some pics and experiement with weights

papanton•11/25/24, 8:35 AM

Is Konya the best for training flux? Where does OneTrainer stand?

Ppapanton Is Konya the best for training flux? Where does OneTrainer stand?

XCogni•11/25/24, 8:58 AM

i'd say for sdxl one trainer is better, and for flux kohya

Zhigarev•11/25/24, 9:12 AM

How do you determine which model or LoRA is better using a GRID? I conducted a test between 10 fine-tuned models, and on all of them, the similarity to a real person looks excellent! What do you rely on? Perhaps you use some special prompts to choose one model?

ZZhigarev How do you determine which model or LoRA is better using a GRID? I conducted a t...

XCogni•11/25/24, 9:33 AM

Depends on what you're aiming for? Flexibility choose the earliest checkpoint that looks like your character, Full likeness choose a later checkpoint.

Zhigarev•11/25/24, 9:49 AM

The thing is, they all look like my character, and it's quite difficult to choose. The question is more about the fact that next time I can use fewer epochs for training, as I understand.

ZZhigarev The thing is, they all look like my character, and it's quite difficult to choos...

XCogni•11/25/24, 9:50 AM

You can always generate samples during training

Zhigarev•11/25/24, 9:50 AM

I think we need to come up with different GRID prompts for the next experiments, such as camera angles and emotions.

Zhigarev•11/25/24, 9:51 AM

something like that

flux-a-comparison-of-camera-angle-prompts-see-first-comment-v0-n3gbofl6obhd1.png.webp

ZZhigarev How do you determine which model or LoRA is better using a GRID? I conducted a t...

Furkan Gözükara SECourses•11/25/24, 10:43 AM

that means training worked perfect but it is subjective and hard

Furkan Gözükara SECourses•11/25/24, 10:43 AM

dont think too much

Badnerle | Can•11/25/24, 12:47 PM

Hey Guys. I have a Question. Im Training on a V100 32GB. I get 10s/it Speed. Is this normal?

Badnerle | Can•11/25/24, 12:47 PM

Batch Size 1.

BBadnerle | Can Hey Guys. I have a Question. Im Training on a V100 32GB. I get 10s/it Speed. Is ...

Furkan Gözükara SECourses•11/25/24, 12:48 PM

yep it is weak https://bizon-tech.com/gpu-benchmarks/NVIDIA-Tesla-V100-vs-NVIDIA-RTX-3090/535vs579?srsltid=AfmBOorxtU8g55liGin88wKhLtkteZ1zAU1b5XibM8hXB2mJjDhZIF4r

BIZON Custom AI Workstation Computers, NVIDIA GPU Servers. AI/ML, Data Science, Deep learning, HPC.

NVIDIA Tesla V100 vs. NVIDIA RTX 3090| GPU Benchmarks for AI/ML, de...

In this article, we are comparing the best graphics cards for deep learning in 2023-2024: NVIDIA RTX 4090 vs RTX 6000, A100, H100 vs RTX 4090

RebelPuppy•11/25/24, 1:00 PM

hey I am trying to get started with ai images of people. confused on howw to best do this

RRebelPuppy hey I am trying to get started with ai images of people. confused on howw to bes...

Furkan Gözükara SECourses•11/25/24, 1:02 PM

eloborate more what you need

Has anyone compared dreambooth 1:1 aspect training vs bucket training yet for Flux? Please let me kn

Similar Threads