Software Engineering Courses (SECourses)•2y ago

<@205854764540362752> upgraded to new xformers 0.27 and torch, getting way better results and smooth

@Furkan Gözükara SECourses upgraded to new xformers 0.27 and torch, getting way better results and smoother results with xformers and training now on kohya

Sergio A.•7/13/24, 7:56 AM

I'm looking for an AI developer to develop my project.
If u are interested, contact me.

TThéo i tried different dataset style, always perfect quality with different clothing ...

Furkan Gözükara SECourses•7/13/24, 8:23 AM

you got some very accurate findings

MMaxivi runpod prices decreased a lot for 6000ada and other gpus, looks like they'll try...

Furkan Gözükara SECourses•7/13/24, 8:25 AM

true i see A6000 dropped from 0.69 to 0.49 but still we get from 0.31 cent on massed compute with our coupon

MMaxivi <@205854764540362752> upgraded to new xformers 0.27 and torch, getting way bette...

Furkan Gözükara SECourses•7/13/24, 8:25 AM

yes xformers newer versions are way better than older ones

TThéo you just follow the onetrainer tutorial from our lord Dr. 1.5 and SDXL works pe...

smoke•7/13/24, 8:36 AM

Thanks a lot for this! Really interesting I appreciate it.

I’ve done a SDXL training with a pretty good dataset I think, on Realvisxl 4, no captions. But the quality wasn’t that great, and the face resemblance was even worse.

https://discord.com/channels/772774097734074388/1261270101244641302/1261336695110897766

I use Comfy as my main tool but the training itself must be a little bit better I feel like, maybe I’m still doing something wrong

Théo•7/13/24, 8:41 AM

don't forget that you need to use the good vae if not baked-in, Hires 1.5/1.7, adetailer capped on resolution 1024x1024 with a denoising strengh between 0.35 to 0.60,

Théo•7/13/24, 8:42 AM

Also i did an update on my stable diffusion and i couldnt anymore use my training face, , everything worked, but not my trained model, suddently turned bad for no reason, i had to reinstall a fresh sb, dunno why but some ppl had the same pb as me, so maybe you have an adetailer version or something that went wrong

smoke•7/13/24, 9:06 AM

I see! I don't think I even used ADetailer for those but I'll have a look. Thanks a lot!

FFurkan Gözükara SECourses true i see A6000 dropped from 0.69 to 0.49 but still we get from 0.31 cent on m...

MaxiviOP•7/13/24, 9:10 AM

hopefully massed compute gets the 6000ada soon for even faster speeds

MMaxivi hopefully massed compute gets the 6000ada soon for even faster speeds

Furkan Gözükara SECourses•7/13/24, 9:19 PM

i totally agree

Furkan Gözükara SECourses•7/13/24, 9:19 PM

@NicB

Xcorpion•7/14/24, 8:09 AM

made 2 loras with completely different 2 datasets, but it gets "blush" as the trigger word.

Xcorpion•7/14/24, 8:10 AM

changing dataset doesnt solve anything

Xcorpion•7/14/24, 8:11 AM

I dont need a trigger word, so how do I disable that?

Xcorpion•7/14/24, 8:14 AM

did enable shuffle caption too

daroth•7/14/24, 8:38 AM

if you already trained it, you cant change the trigger word

XXcorpion did enable shuffle caption too

Furkan Gözükara SECourses•7/14/24, 8:58 AM

it is about your captions

Furkan Gözükara SECourses•7/14/24, 8:58 AM

you need to understand how captions work

FFurkan Gözükara SECourses it is about your captions

Xcorpion•7/14/24, 9:37 AM

hmm used this to captioning, thats it. I mean I made 5loras prev, those are fine.

TheToday•7/14/24, 2:47 PM

If I already have a trained model but I want to add more poses/expressions, would it be correct to take that already trained checkpoint and try only 3 or 4 photos with those expressions? What is your opinion? Because doing it again costs me more money, and this way I save time

yoziv•7/15/24, 2:29 AM

Prepping a dataset right now of a human subject and most of the quality photos I have are group shots. Is there a way I can modify/psd the group shots to work for training purposes?

Superman•7/15/24, 7:20 AM

Can someone please tell what base model should i use to train a face like this (half cartoon 3d render and half human like)

I was made using faceswap with an cartoon face

SSuperman Can someone please tell what base model should i use to train a face like this (...

MaxiviOP•7/15/24, 8:20 AM

https://civitai.com/models/216439/artium

Artium - v2.0 | Stable Diffusion Checkpoint | Civitai

A R T I U M An offshoot from my OMNIUM model, geared more towards an illustrative and painterly style, still very much infused with Omnium's DNA in...

TTheToday If I already have a trained model but I want to add more poses/expressions, woul...

Furkan Gözükara SECourses•7/15/24, 8:35 AM

i would improve dataset and do training again

Yyoziv Prepping a dataset right now of a human subject and most of the quality photos I...

Furkan Gözükara SECourses•7/15/24, 8:36 AM

i would avoid

Furkan Gözükara SECourses•7/15/24, 8:36 AM

because you will get mixed faces

SSuperman Can someone please tell what base model should i use to train a face like this (...

Furkan Gözükara SECourses•7/15/24, 8:36 AM

i would download best models from civitiai

Furkan Gözükara SECourses•7/15/24, 8:36 AM

and test my prompt

Furkan Gözükara SECourses•7/15/24, 8:36 AM

and see which one yields best ones

Furkan Gözükara SECourses•7/15/24, 8:36 AM

for generic prompts

J2_Nedo•7/15/24, 10:08 AM

Question, I have had very good results with Onetrainer SD1.5 finetuning and LoRa extraction, now I am trying it with SDXL models, finetuning works very well,
but when I extract LoRa the similarity of the person is much weaker, this is not the case with the SD1.5 extracted LoRAs, do I just have to train longer here ?

daroth•7/15/24, 10:17 AM

btw, how big of a quality difference are we talking about when it comes to direct lora draining vs full checkpoint training + extraction?

Ddaroth btw, how big of a quality difference are we talking about when it comes to direc...

J2_Nedo•7/15/24, 10:28 AM

at least with my datasets i get the best results with full checkpoint training + extraction. How big the difference? i can see it, if extraction is 10 then i would give a 6 or 7

Fireart•7/15/24, 11:36 AM

Hello! I would like to ask for some advice. How can I achieve the best settings for training SDXL Lora on OneTrainer, and what should I pay attention to? The training set consists of approximately 80-150 images of artist’s works with inconsistent aspect ratios. Thank you!

MMaxivi hopefully massed compute gets the 6000ada soon for even faster speeds

NicB•7/15/24, 2:28 PM

The L40 & L40S are very similar to the 6000 ADA. We have a few of those available right now.

The L40 & L40S have a better server CPU and Ram as well.

NNicB The L40 & L40S are very similar to the 6000 ADA. We have a few of those availabl...

MaxiviOP•7/15/24, 4:07 PM

i think in terms of pricing with runpod both the l40 and 6000ada are quite cheap 79cents per hour

but on massed compute its 1.15 for the l40

MMaxivi i think in terms of pricing with runpod both the l40 and 6000ada are quite cheap...

NicB•7/15/24, 9:25 PM

Apples to apples would be secure cloud pricing on runpod.