Software Engineering Courses (SECourses)•3y ago

I’m not sure, I think they want us to test just interference first. I’d suspect more features to c

I’m not sure, I think they want us to test just interference first.
I’d suspect more features to come later. But check it out lmk

HaashimOP•3/25/23, 1:53 AM

That’s why I like the cloud though. You can run multiple machines and still do your computing

ZZono50 In captions they say to put what you don't want the ai to learn, is that true fo...

Furkan Gözükara SECourses•3/25/23, 1:54 AM

filewords is captions

n8hawk_•3/25/23, 1:54 AM

stable horde seems to sth similar

HHaashim I’m not sure, I think they want us to test just interference first. I’d suspec...

n8hawk_•3/25/23, 2:14 AM

works!

Zono50•3/25/23, 2:17 AM

when would you want to use [filewords]

n8hawk_•3/25/23, 2:23 AM

that is very nice- it simply works while my local grafikcards is doing some dreambooth training.

ZZono50 when would you want to use [filewords]

Furkan Gözükara SECourses•3/25/23, 2:37 AM

when you have so many good images for each token. like improve castles horses cats etc

Nn8hawk_that is very nice- it simply works while my local grafikcards is doing some drea...

Furkan Gözükara SECourses•3/25/23, 2:38 AM

probably free for a certain time

Furkan Gözükara SECourses•3/25/23, 2:38 AM

how many limit it has you checked?

n8hawk_•3/25/23, 2:39 AM

you mean dreamstudio.ai?

n8hawk_•3/25/23, 2:40 AM

i already had an account for 6 month or longer and the info is
"103,3 credits

Approximately 516 images with default settings"

n8hawk_•3/25/23, 2:41 AM

with respect to the 50.000 images i have created in the last month that will not help me for long

FFurkan Gözükara SECourses probably free for a certain time

n8hawk_•3/25/23, 2:44 AM

is it correct that I have to double the training steps per image, when i halften the learning rate?

n8hawk_•3/25/23, 2:45 AM

when dreambooth training?

Nn8hawk_is it correct that I have to double the training steps per image, when i halften...

Furkan Gözükara SECourses•3/25/23, 2:50 AM

i cant say there is such deterministic correlation

n8hawk_•3/25/23, 2:52 AM

ok, good to know.thank you! i will retire now- it is 4am

Furkan Gözükara SECourses•3/25/23, 3:00 AM

ye i am gonna sleep now

Furkan Gözükara SECourses•3/25/23, 3:00 AM

6 am

Furkan Gözükara SECourses•3/25/23, 3:00 AM

hopefully see u tomorrow

t7ds001•3/25/23, 3:30 AM

I wanted to know if your lora training tutorial here covers making sure this lora training doesn't break other models that have trained tokens in them? https://www.youtube.com/watch?v=mfaqqL5yOO4 I notice with textual inversion, merging checkpoints or loras to another model results in low quality. I tried changing lora dim/alpha as well and the quality was not good when using my other trained tokens from other models at the same time of using the lora I properly trained or extracted and I made sure to lower the lora value and cfg value. I did lora training with kohya. So I was wondering if webui dreambooth extension with lora actually has a different result that keeps the quality of other models intact with tokens we have trained into them. When I did dreambooth with shivam's repo, not dreambooth extension, but his actual repo, my training is fine individually but as soon as I train both training sets together in one model, the quality with using them at the same time is not good. @Furkan Gözükara SECourses

YouTubeSECourses - Software Engineering Courses

How To Do Stable Diffusion LORA Training By Using Web UI On Differe...

Our Discord : https://discord.gg/HbqgGaZVmr. Ultimate guide to the LoRA training. If I have been of assistance to you and you would like to show your support for my work, please consider becoming a patron on

https://www.patreon.com/SECourses

Playlist of Stable Diffusion Tutorials, #Automatic1111 and Google Colab Guides, DreamBooth, Textual In...

How To Do Stable Diffusion LORA Training By Using Web UI On Differe...

KJ•3/25/23, 5:50 AM

is it still necessary to crop images by 512x512 for dreambooth? how much results will differ without cropping and using enable buckets

buriedanimal•3/25/23, 5:59 AM

does anyone how you would merge two checkpoints trained on different subject keywords to combine them? they're two different people i want to combine.

KKJ is it still necessary to crop images by 512x512 for dreambooth? how much results...

Stevens•3/25/23, 6:00 AM

Everything I have heard and read says you have to make them 512 x 512 and close up of the face

SStevens Everything I have heard and read says you have to make them 512 x 512 and close ...

KJ•3/25/23, 6:00 AM

it is for style, not a person

Bburiedanimal does anyone how you would merge two checkpoints trained on different subject key...

Stevens•3/25/23, 6:02 AM

I don't think you can merge the two people into one person but you can merge the two checkpoints on the merge tab which would make both people available in one checkpoint. Hopefully I did not misunderstand you!

KKJ it is for style, not a person

Stevens•3/25/23, 6:02 AM

Gotcha. Sorry for the mistake. I have no experience with styles yet. but there are a bunch of cool people so I am sure someone can help

SStevens I don't think you can merge the two people into one person but you can merge the...

buriedanimal•3/25/23, 6:05 AM

yeah, i know i can merge them there. I guess i should have trained them using the same instance token/keyword.

buriedanimal•3/25/23, 6:07 AM

oh wait i needed to rtfm

Stevens•3/25/23, 6:12 AM

lmk how it turns out because that seems like it would be good for definitely not anything shady

FORTUNE VARO•3/25/23, 7:05 AM

Hi! I have lots of very good img2img images I did looking for good prompts for an animation in Deforum so, I now that you can set an Init image and that´s great! My question is the following: Is there a way to set or keyframe different images as inits every X frames? To tell the proces to "activate" a reference y an specific time?

SStevens lmk how it turns out because that seems like it would be good for definitely not...

buriedanimal•3/25/23, 7:56 AM

ok it turns out it's nightmarish unless you make sure all of them look okay on the same samplers, one of my models couldn't merge DDIM or PLMS. It doesn't look better than a chkpt + LORA.

Stevens•3/25/23, 7:58 AM

roger that. i mean, you are light years ahead of me, but I understand. I am still trying to make one embed work but Doc says I need better pics.

VogBy•3/25/23, 9:15 AM

Hello, I have some trouble with installing Stable diffusion Web UI

VogBy•3/25/23, 9:16 AM

I'm currently looking at this video : https://www.youtube.com/watch?v=AZg6vzWHOTA

YouTubeSECourses - Software Engineering Courses

Easiest Way to Install & Run Stable Diffusion Web UI on PC by Using...

Our Discord : https://discord.gg/HbqgGaZVmr. How to download and install Stable Diffusion Automatic1111 Web UI interface. If I have been of assistance to you and you would like to show your support for my work, please consider becoming a patron on

https://www.patreon.com/SECourses

Playlist of Stable Diffusion Tutorials, Automatic1111 and Goog...

Easiest Way to Install & Run Stable Diffusion Web UI on PC by Using...

VogBy•3/25/23, 9:17 AM

and when I downloaded the .exe file and launched it, it didn't create a git file. And for more, I can't start the file

Tt7ds001 I wanted to know if your lora training tutorial here covers making sure this lor...

Furkan Gözükara SECourses•3/25/23, 10:24 AM

well i havent tested yet with lora. i am planning hopefully soon. but i think newest version of dreambooth extension is much better than before .

KKJ is it still necessary to crop images by 512x512 for dreambooth? how much results...

Furkan Gözükara SECourses•3/25/23, 10:24 AM

no you can have different dimensions but make sure that you have increased max resolution

Bburiedanimal does anyone how you would merge two checkpoints trained on different subject key...

Furkan Gözükara SECourses•3/25/23, 10:25 AM

i think you can use this method : https://youtu.be/s25hcW4zq4M . if they are same token they will bleed. you have to train different tokens

YouTubeSECourses - Software Engineering Courses

How to Inject Your Trained Subject e.g. Your Face Into Any Custom S...

Our Discord : https://discord.gg/HbqgGaZVmr. In this video, I am explaining how to extract the taught subject information from your trained model and inject into any new custom Stable Diffusion model. If I have been of assistance to you and you would like to show your support for my work, please consider becoming a patron on

https://www.patreo...

How to Inject Your Trained Subject e.g. Your Face Into Any Custom S...

FFORTUNE VARO Hi! I have lots of very good img2img images I did looking for good prompts for a...

Furkan Gözükara SECourses•3/25/23, 10:27 AM

no idea for that atm. i am testing img2img alternative test

VVogBy and when I downloaded the .exe file and launched it, it didn't create a git file...

Furkan Gözükara SECourses•3/25/23, 10:27 AM

welcome. when you type python to your cmd what do you see

jack kaw•3/25/23, 10:42 AM

Launching Web UI with arguments: --no-half --xformers
C:\stable-diffusion-webui\venv\lib\site-packages\torch\cuda__init__.py:88: UserWarning: CUDA initialization: CUDA unknown error - this may be due to an incorrectly set up environment, e.g. changing env variable CUDA_VISIBLE_DEVICES after program start. Setting the available devices to be zero. (Triggered internally at ..\c10\cuda\CUDAFunctions.cpp:109.)
return torch._C._cuda_getDeviceCount() > 0
Warning: caught exception 'CUDA unknown error - this may be due to an incorrectly set up environment, e.g. changing env variable CUDA_VISIBLE_DEVICES after program start. Setting the available devices to be zero.', memory monitor disabled
Loading weights [27a4ac756c] from C:\stable-diffusion-webui\models\Stable-diffusion\SD15NewVAEpruned.ckpt
Creating model from config: C:\stable-diffusion-webui\configs\v1-inference.yaml
LatentDiffusion: Running in eps-prediction mode
DiffusionWrapper has 859.52 M params.
loading stable diffusion model: RuntimeError
Traceback (most recent call last):

jack kaw•3/25/23, 10:42 AM

File "C:\stable-diffusion-webui\webui.py", line 136, in initialize
modules.sd_models.load_model()
File "C:\stable-diffusion-webui\modules\sd_models.py", line 445, in load_model
sd_hijack.model_hijack.hijack(sd_model)
File "C:\stable-diffusion-webui\modules\sd_hijack.py", line 172, in hijack
self.optimization_method = apply_optimizations()
File "C:\stable-diffusion-webui\modules\sd_hijack.py", line 42, in apply_optimizations
if cmd_opts.force_enable_xformers or (cmd_opts.xformers and shared.xformers_available and torch.version.cuda and (6, 0) <= torch.cuda.get_device_capability(shared.device) <= (9, 0)):
File "C:\stable-diffusion-webui\venv\lib\site-packages\torch\cuda__init.py", line 357, in get_device_capability
prop = get_device_properties(device)
File "C:\stable-diffusion-webui\venv\lib\site-packages\torch\cuda__init.py", line 371, in get_device_properties
_lazy_init() # will define _get_device_properties
File "C:\stable-diffusion-webui\venv\lib\site-packages\torch\cuda__init__.py", line 229, in _lazy_init
torch._C._cuda_init()
RuntimeError: CUDA unknown error - this may be due to an incorrectly set up environment, e.g. changing env variable CUDA_VISIBLE_DEVICES after program start. Setting the available devices to be zero.

Stable diffusion model failed to load, exiting
Press any key to continue . . .

jack kaw•3/25/23, 10:43 AM

-.-

jack kaw•3/25/23, 10:43 AM

Zono50•3/25/23, 10:45 AM

Hey doc, my trained model i made yesterday, if i literally do any prompt other than photo of ohwx woman <loratag>, it doesn't look like her anymore. What can I do to fix that?

Zono50•3/25/23, 10:46 AM

this is her as lara croft from tomb raider as the only prompt

Zono50•3/25/23, 10:47 AM

jack kaw•3/25/23, 10:48 AM

Jjack kaw Launching Web UI with arguments: --no-half --xformers C:\stable-diffusion-webui\...

Furkan Gözükara SECourses•3/25/23, 12:42 PM

what is your graphic card

ZZono50 Hey doc, my trained model i made yesterday, if i literally do any prompt other t...

Furkan Gözükara SECourses•3/25/23, 12:43 PM

you tried different cfg and prompt strength

I’m not sure, I think they want us to test just interference first. I’d suspect more features to c

Similar Threads