Software Engineering Courses (SECourses)•3y ago

Good morning 3 days ago I was able to print 2048x2048 images with 4060 8gb vram, but now I can not

Good morning
3 days ago I was able to print 2048x2048 images with 4060 8gb vram, but now I can not print.
If idle vram usage can help me in this case, thank you very much in advance.

DoluBura- esatOP•12/5/23, 5:21 AM

RuntimeError: Not enough memory, use lower resolution (max approx. 1600x1600). Need: 6.0GB free, Have:4.2GB free

DoluBura- esatOP•12/5/23, 5:21 AM

DoluBura- esatOP•12/5/23, 5:23 AM

With no models or lora installed, vram usage is 2.8 vram.

DoluBura- esatOP•12/5/23, 5:24 AM

My whole problem now is what is the cause of this and how can I use it as an effect.

DoluBura- esatOP•12/5/23, 5:27 AM

Belirttiğim gibi bütün sorunum bu ingilzce yazdım sunucuda hep öyle konuşulmuş diye belki çözümünü bilen vardır.

FFurkan Gözükara SECourses i may be sleeping but can commet tomorrow hopefully

shortdwarf•12/5/23, 6:34 AM

Sorry for taking so long to get back to you. Upon looking into it more online, it seems it is an error with if your first engine is not default. After youve made your first with default settings it will allegedly work. Ill edit this comment if it works.
EDIT: This was the solution.

FFurkan Gözükara SECourses i may be sleeping but can commet tomorrow hopefully

shortdwarf•12/5/23, 7:14 AM

Any idea how to prevent this:
UNSUPPORTED_STATESkipping tactic 0 due to insufficient memory on requested size of 7461533184 detected for tactic 0x00
Attempting to do 1536x1024. I have 12gb vram, do you think medvram could fix it?

Lebrun•12/5/23, 7:53 AM

is it possible to use sd 1.5 based model with sdxl refiner

nissanle•12/5/23, 8:26 AM

Hello Furkan,
I hope you are well.
As you may remember I have meticulously built both A1111 and Kohya_ss platforms according to your great tutorials.
With my last LORA training I unfortunately get broken and corrupted images like you see in the enclosed examples.
Can you please tell me what should I do to rectify this issue?
Thank you.

FFurkan Gözükara SECourses magic-animate

Elisoli•12/5/23, 9:17 AM

The question i have with magic animate - if you use a photo of yourself and than the pre made dancing - would the resault's face looks like yours? or is it autogenerated and wouldnt look like it at all?

Nnissanle Hello Furkan, I hope you are well. As you may remember I have meticulously built...

DR.Siva•12/5/23, 9:53 AM

I am getting the same results , i was able to avoid by

check resolution 1024x1024 ( start with this res)
Decrease steps to 30
Use 2M karras sampler
Use some other models (ProtovisionXL,NightvisionXL)

DR.Siva•12/5/23, 9:55 AM

euler a does produce these, but it has to do with sampling step as well. in my case the last step creates these artificats. Also which lora output number are you using ? For me somewhere from 11-15 gives decent resuilts

Deleted User•12/5/23, 10:26 AM

How to extract lora from dreambooth model thank you

DDR.Siva euler a does produce these, but it has to do with sampling step as well. in my c...

nissanle•12/5/23, 11:50 AM

DR.Siva thank you so much for your kind reply. I’ll implement your guiding points and will let you know of the outcome.
What do you mean by lora output number? In my training parameters I used 6 epochs so I got 6 models, each of 147,571 KB.

DR.Siva•12/5/23, 11:54 AM

You trained dreambooth?

DR.Siva•12/5/23, 11:55 AM

This indicates overtraining

andrsoe•12/5/23, 12:36 PM

Hello, after much frustration trying to work with a1111 and controlnets, using a1111 is an API backend service i come to you for help!

I just want someone who can link me json payload they send to a1111 running sdxl where the output makes sense compared to settings things up in the GUI:
-json payload to a1111 for any sdxl checkpoint w. controlnet they use that works, img2img and or txt2img
-also understanding in the api why i cant upload an image for controlNet for txt2img, but only in img2img

-> My problem is when using controlnet in my api call its like im doing regular img2img without the controlnet, and in the api call i cant upload my controlNet. In a1111 i can setup settings for perfect image generation for txt2img, w controlnet, but cant get the same results when running in --api mode, sending the json payload :S

Aandrsoe Hello, after much frustration trying to work with a1111 and controlnets, using a...

lucy•12/5/23, 12:55 PM

a11 is horrible for API usage especially with third party plugins injecting, comfy UI albeit more messy to setup at first if youre not familiar, allows for much easier flow and has native API via dev mode

DDR.Siva You trained dreambooth?

nissanle•12/5/23, 12:55 PM

I trained lora with kohya_ss.

Llucy a11 is horrible for API usage especially with third party plugins injecting, com...

andrsoe•12/5/23, 12:58 PM

are there good guides for getting things started? Like heavy on the developer front? and i mean in general for cloud gpu / api shit / workflow everything. im looking for a hub/journy of learning which is why my question is maybe annoying, but thanks for input about comfy and api, usage.
-> thoughts on where to start with comfy api mode, or i will just google and pray result number 1 is good. But thanks either way.

Aandrsoe are there good guides for getting things started? Like heavy on the developer fr...

lucy•12/5/23, 12:59 PM

https://github.com/comfyanonymous/ComfyUI/blob/master/script_examples/basic_api_example.py

GitHub

ComfyUI/script_examples/basic_api_example.py at master · comfyanony...

The most powerful and modular stable diffusion GUI with a graph/nodes interface. - comfyanonymous/ComfyUI

lucy•12/5/23, 1:00 PM

you'd setup the flow in comfy the way you want, then dump it in api format, and use your favorite language of choice to send a json

andrsoe•12/5/23, 1:09 PM

Anyway, last prayer for someone who uses a1111 with --api and can send me the payload they use for successfull image generation:
I just want someone who can link me json payload they send to a1111 running sdxl where the output makes sense compared to settings things up in the GUI:
-json payload to a1111 for any sdxl checkpoint w. controlnet they use that works, img2img and or txt2img
-also understanding in the api why i cant upload an image for controlNet for txt2img, but only in img2img

-> feel free to delete, im just begging here with mouth open

Aandrsoe Anyway, last prayer for someone who uses a1111 with --api and can send me the pa...

Benjamin•12/5/23, 1:18 PM

there is a plugin that will show the json for api call after image generation

Benjamin•12/5/23, 1:18 PM

I need to look up the name

Benjamin•12/5/23, 1:21 PM

ah it is even mentioned in https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/API

GitHub

API

Stable Diffusion web UI. Contribute to AUTOMATIC1111/stable-diffusion-webui development by creating an account on GitHub.

Benjamin•12/5/23, 1:22 PM

this one https://github.com/huchenlei/sd-webui-api-payload-display

GitHub

GitHub - huchenlei/sd-webui-api-payload-display: Display the corres...

Display the corresponding API payload after each generation on WebUI - GitHub - huchenlei/sd-webui-api-payload-display: Display the corresponding API payload after each generation on WebUI

Nnissanle I trained lora with kohya_ss.

DR.Siva•12/5/23, 1:28 PM

Just 6 epoch?
I have 16

BBenjamin ah it is even mentioned in https://github.com/AUTOMATIC1111/stable-diffusion-web...

andrsoe•12/5/23, 1:30 PM

thanks a lot, but damn i feel a bit dumb rn, but fuck me thanks

Aandrsoe thanks a lot, but damn i feel a bit dumb rn, but fuck me thanks

Benjamin•12/5/23, 1:31 PM

no worries, enjoy

JJ.does anyone know how to do fooocus+ get an api? I read some github issue where i...

Furkan Gözükara SECourses•12/5/23, 1:37 PM

are you sure fooocus has API version? i dont see on readme

Aaliattaran @Dr. Furkan Gözükara hi, which tier should i join to have the face swap code set...

Furkan Gözükara SECourses•12/5/23, 1:38 PM

even lowest tier opens all. but if you register higher thank you

DDoluBura- esat Good morning 3 days ago I was able to print 2048x2048 images with 4060 8gb vram...

Furkan Gözükara SECourses•12/5/23, 1:38 PM

are you using rc version? 1.7.

Sshortdwarf Any idea how to prevent this: UNSUPPORTED_STATESkipping tactic 0 due to insuffi...

Furkan Gözükara SECourses•12/5/23, 1:40 PM

yes medvram certainly can fix

LLebrun is it possible to use sd 1.5 based model with sdxl refiner

Furkan Gözükara SECourses•12/5/23, 1:41 PM

on foocus you are able to use i think

Furkan Gözükara SECourses•12/5/23, 1:41 PM

but havent tried on automatic1111

Furkan Gözükara SECourses•12/5/23, 1:41 PM

you should be

Nnissanle Hello Furkan, I hope you are well. As you may remember I have meticulously built...

Furkan Gözükara SECourses•12/5/23, 1:41 PM

looks like VAE or resolution mismatch

EElisoli The question i have with magic animate - if you use a photo of yourself and than...

Furkan Gözükara SECourses•12/5/23, 1:41 PM

should be like yours. you can try

Furkan Gözükara SECourses•12/5/23, 1:41 PM

auto installer working

Furkan Gözükara SECourses•12/5/23, 1:42 PM

i am trying to make auto installer for dense pose generator

Furkan Gözükara SECourses•12/5/23, 1:42 PM

spent like 4 hours todaty

DDeleted User How to extract lora from dreambooth model thank you

Furkan Gözükara SECourses•12/5/23, 1:42 PM

use tools of Kohya GUI so easy

Aandrsoe Hello, after much frustration trying to work with a1111 and controlnets, using a...

Furkan Gözükara SECourses•12/5/23, 1:42 PM

this is a pro job. you probably need to pay a lot for a programmer

FFurkan Gözükara SECourses i am trying to make auto installer for dense pose generator

lucy•12/5/23, 1:45 PM

magic animate seems to have a different color palette than the ones in the linked detectron repo also, or maybe it's just the fact the background is that shade of purple?

lucy•12/5/23, 1:47 PM

would need to also split a video into its images with ffmpeg in a previous step, then apply_net and then reassemble the output images it did looks like

Llucy magic animate seems to have a different color palette than the ones in the linke...

Furkan Gözükara SECourses•12/5/23, 1:48 PM

the problem is i still couldnt make detectron2 work

freaking hard

Furkan Gözükara SECourses•12/5/23, 1:48 PM

https://github.com/facebookresearch/detectron2/issues/5170

GitHub

No object named 'DensePoseROIHeads' found in 'ROI_HEADS' registry! ...

I have installed and working. But there is literally 0 documentation and examples. in the same folder densepose_rcnn_R_101_FPN_DL_s1x.yaml located in the same folder Base-DensePose-RCNN-FPN.yaml lo...

FFurkan Gözükara SECourses looks like VAE or resolution mismatch

nissanle•12/5/23, 1:49 PM

Hello Furkan, DR. Siva replied the following to me:
I am getting the same results , i was able to avoid by

check resolution 1024x1024 ( start with this res)
Decrease steps to 30
Use 2M karras sampler
Use some other models (ProtovisionXL,NightvisionXL)
euler a does produce these, but it has to do with sampling step as well. in my case the last step creates these artificats. Also which lora output number are you using ? For me somewhere from 11-15 gives decent resuilts

Nnissanle Hello Furkan, DR. Siva replied the following to me: I am getting the same result...

Furkan Gözükara SECourses•12/5/23, 1:52 PM

good info

Good morning 3 days ago I was able to print 2048x2048 images with 4060 8gb vram, but now I can not

Similar Threads