Software Engineering Courses (SECourses)•3y ago

https://github.com/Gourieff/sd-webui-reactor

GitHub

GitHub - Gourieff/sd-webui-reactor: Fast and Simple Face Swap Exten...

Fast and Simple Face Swap Extension for StableDiffusion WebUI (A1111, SD.Next, Cagliostro) - GitHub - Gourieff/sd-webui-reactor: Fast and Simple Face Swap Extension for StableDiffusion WebUI (A1111...

king•10/6/23, 2:14 PM

Hi @here

SSandstormleader Click to see attachment

LEGACIES•10/6/23, 2:16 PM

how much training steps did you use?

Sandstormleader•10/6/23, 2:16 PM

150

LEGACIES•10/6/23, 2:20 PM

how much images is there in your dataset?

Sandstormleader•10/6/23, 2:21 PM

this one above about 26

Sandstormleader•10/6/23, 2:21 PM

but a lot 10 were probably the same image cropped closer to face

LEGACIES•10/6/23, 2:22 PM

i will try with 20 images first

Sandstormleader•10/6/23, 2:22 PM

so i think i'm just going to set up a white backdrop for myself this afternoon

Sandstormleader•10/6/23, 2:24 PM

i also do not downscale my images

Sandstormleader•10/6/23, 2:25 PM

so that first image was 2581x3872

Sandstormleader•10/6/23, 2:32 PM

@Dr. Furkan Gözükara have you tested results of data photos having a beard vs no beard?

Sandstormleader•10/6/23, 2:38 PM

wondering if i need to shave lol

rafraf•10/6/23, 2:58 PM

Hey @Dr. Furkan Gözükara got amazing results for 12 image character LoRA using AdamW8bit 8000 steps--testing different batch sizes quality vs speed of training. Did you test LoRA Adafactor settings or only Full Finetune? I tried LoRA Adafactor with your settings with LR 0.0004 and it learned very little, color discoloration etc.. Can you point in right direction for LoRA--Adafactor?

rafraf•10/6/23, 3:41 PM

I see you are checking Adafactor SDXL TE problem on Issue #855. Does this mean Adafactor is broken at the moment, in light of TE? I thought it was beneficial to exclude SDXL Text Encoder from learning? Is Dadaptation a better option?

riccardo•10/6/23, 3:54 PM

can someone explain why google colab gives me this error:
ImportError Traceback (most recent call last)
<ipython-input-7-fdf29cbd2fa1> in <cell line: 22>()
20 import io
21 import math
---> 22 import timm
23 from IPython import display
24 import lpips

11 frames
/usr/local/lib/python3.10/dist-packages/PIL/ImageFont.py in <module>
32
33 from . import Image
---> 34 from ._util import isDirectory, isPath
35
36 LAYOUT_BASIC = 0

ImportError: cannot import name 'isDirectory' from 'PIL._util' (/usr/local/lib/python3.10/dist-packages/PIL/_util.py)

---------------------------------------------------------------------------
NOTE: If your import is failing due to a missing package, you can
manually install dependencies using either !pip or !apt.

To view examples of installing some common dependencies, click the
"Open Examples" button below.

Jehex•10/6/23, 4:24 PM

Hello guy's, How can I run 2 instances of SD ( a1111 ) at the same time ?

Mmikemenders However, there is a striking difference between SD and RealisticVision, and betw...

EyeSpyBekoAI•10/6/23, 6:05 PM

Hey @mikemenders , try RV5 (not 5.1) with the 8400 VAE. I personally had better luck. And yes, SD1.5 vs. RV models are worlds apart.

EyeSpyBekoAI•10/6/23, 6:06 PM

Regardless though your pics look great!

Sandstormleader•10/6/23, 8:29 PM

so the --share command thing is broken i believe

Sandstormleader•10/6/23, 8:33 PM

Could not create share link. Please check your internet connection or our status page: https://status.gradio.app
Startup time: 13.1s (import torch: 1.2s, import gradio: 0.5s, import ldm: 0.3s, other imports: 0.8s, load scripts: 4.6s, create ui: 0.3s, gradio launch: 5.4s).

Mmikemenders https://github.com/Gourieff/sd-webui-reactor

EyeSpyBekoAI•10/6/23, 9:23 PM

Hey @mikemenders , this is interesting. Do you know the difference between this vs. adetailer, or have you tested them against each other?

JJehex Hello guy's, How can I run 2 instances of SD ( a1111 ) at the same time ?

thesilvermoth•10/6/23, 9:26 PM

I also want to know this. Do I need different SD folders?

SSandstormleader so the --share command thing is broken i believe

thesilvermoth•10/6/23, 9:27 PM

It will only work if SD is on local. You can't use that command on RunPod or Kaggle

Sandstormleader•10/6/23, 9:36 PM

I only use local

Sandstormleader•10/6/23, 9:37 PM

Do I need an extension or something as well?

PIXLE•10/6/23, 9:45 PM

Question for any of you: I have trained LORAs for my wife, myself, and my two kids. Whats the best way to compose a family portrait or group photo? Any good tutorials out there? I haven't found much.

Sandstormleader•10/6/23, 9:48 PM

i thought about doing something lie that but my approach was going to train multiple concepts within the same model with dreambooth

PPIXLE Question for any of you: I have trained LORAs for my wife, myself, and my two ki...

EyeSpyBekoAI•10/6/23, 9:54 PM

@xPIXELPAPIx you'll need to use the Loras in a single prompt. So for example it'll be something like "husband:lora" sitting next to "wife:lora" with "son:lora" chasing "daughter:lora" in a living room. Something along those lines. I'm not too experienced with complicated Loras myself but it should be something along those lines and there are a few youtube videos out there of people using multiple loras.

Kking Hi @here

Furkan Gözükara SECourses•10/6/23, 10:10 PM

SSandstormleader @Dr. Furkan Gözükara have you tested results of data photos having a beard vs no...

Furkan Gözükara SECourses•10/6/23, 10:10 PM

i dont think you need

Furkan Gözükara SECourses•10/6/23, 10:11 PM

i can get shaved photo :d

Rrafraf Hey @Dr. Furkan Gözükara got amazing results for 12 image character LoRA using A...

Furkan Gözükara SECourses•10/6/23, 10:16 PM

i will explore lora adafactor later. currently finding best hyper params for dreambooth.

Rrafraf I see you are checking Adafactor SDXL TE problem on Issue #855. Does this mean A...

Furkan Gözükara SECourses•10/6/23, 10:17 PM

i got super interesting finding with TE still researching. but we really need different LR for TE : https://twitter.com/GozukaraFurkan/status/1710416135747748150

Furkan Gözükara (@GozukaraFurkan)

SDXL text encoder turns out to be very strong and becomes so quickly overtrained

Twitter

•

10/6/23, 10:05 PM

Rriccardo can someone explain why google colab gives me this error: ImportError ...

Furkan Gözükara SECourses•10/6/23, 10:18 PM

which colab?

JJehex Hello guy's, How can I run 2 instances of SD ( a1111 ) at the same time ?

Furkan Gözükara SECourses•10/6/23, 10:18 PM

yes

FFurkan Gözükara SECourses i got super interesting finding with TE still researching. but we really need di...

rafraf•10/6/23, 10:18 PM

yea, i saw the thread. Sounds like very difficult to test for TE? Have you got some idea about it? I thought there being two TE's made it impossible practically

Furkan Gözükara SECourses•10/6/23, 10:18 PM

but it really doesnt worth unless you have 2 gpu

Rrafraf yea, i saw the thread. Sounds like very difficult to test for TE? Have you got s...

Furkan Gözükara SECourses•10/6/23, 10:18 PM

i dont know still searching

SSandstormleader Could not create share link. Please check your internet connection or our status...

Furkan Gözükara SECourses•10/6/23, 10:19 PM

ye it wasnt working yesterday

Furkan Gözükara SECourses•10/6/23, 10:19 PM

and today for a while

rafraf•10/6/23, 10:19 PM

I am testing AdamW8bit quality with batch sizes 1, 2, 4, 6. I am seeing what is the best sweetspot for quality vs speed of training

PPIXLE Question for any of you: I have trained LORAs for my wife, myself, and my two ki...

Furkan Gözükara SECourses•10/6/23, 10:19 PM

you can try latent couple extension

Rrafraf I am testing AdamW8bit quality with batch sizes 1, 2, 4, 6. I am seeing what is ...

Furkan Gözükara SECourses•10/6/23, 10:19 PM

yep

Furkan Gözükara SECourses•10/6/23, 10:19 PM

this is necessary for companies

Furkan Gözükara SECourses•10/6/23, 10:19 PM

i will also research this for a company

SSandstormleader @Dr. Furkan Gözükara have you tested results of data photos having a beard vs no...

Furkan Gözükara SECourses•10/6/23, 10:23 PM

it adds beard hard

i tested not easy to get without a beard

rafraf•10/6/23, 10:25 PM

i got pretty impressive results with batch 2, which drove SDXL adamW8bit training by by 30% and bs 4 by 60%. For me it's somewhere in there. Need to check flexibility, but i am sure there are other experimental variables to test. Here's dataset of 12 images, vs xy plot with batch 2 up to 9 epochs (5000 steps), 5 hours vs 9 hours batch size 1

xyz_grid-0007-20231006232009-__lora_4du7_0103-lora_adamw8bit_40rep2bs-dim128-alpha64-000001_0.8___closeup_portrait_of_4du7_woman_green_studio_background_st.jpg

rafraf•10/6/23, 10:26 PM

interesting the level of variance in the dataset gets picked up in dim 128/alpha64. I guess the ultimate test will come when i run this LoRA on a Art Style model

REEM•10/6/23, 10:28 PM

@Dr. Furkan Gözükara Hello, I'm trying to train LoRa on objects but it's not working. What's the reason? All the sources I saw were for faces

RREEM @Dr. Furkan Gözükara Hello, I'm trying to train LoRa on objects but it's not wor...

Furkan Gözükara SECourses•10/6/23, 10:30 PM

well it works but object training is a different methodology

https://github.com/Gourieff/sd-webui-reactor

Similar Threads

Similar Threads

Similar Threads