If you want to improve generation performance, and reduce to 5-7 steps while maintaining quality, I recommend using a Lora model ( lcm-lora-sdv1-5 ) You have to add "Add Network to Prompt": in UI settings you can add SD Lora to appear
thank you thank you thank you, I am now able with my little Ryzen 5 and integrated GPU Vega 11 to use it one hundred per cent, before to do a 512x512 Model 1.5 it took 7 minutes, now with Directml it takes ABOUT 1 MINUTE!!! You are great!!!😇😁🤩
Try this before anything else: Uninstall the python versions from your system and install only 3.10.6 (Add to Patch) This will also ensure that there are no errors during the process. Other versions with the Patch get in the way (at least it works for me)
@@tongtongrizhi bro can you make tutorial on how to add reActor to the extentions . i'm asking you to as you make as the tutorial that you make actually work actually work.
Not all stable diffusion sources are the same. Where did you get your zip from? The hash code for git checkout only works specifically on Stable Diffusion web UI with DirectML by lshqqytiger. Any other source other than this won't work. I suggest you follow every steps in the video without skipping any of them. Hope it helps
@@tongtongrizhi I got it directly from your link (by lshqqytiger), could not clone it since I have an unstable network. I had to resort to the Zip file provided in the green code button. But now the "Git checkout code" does not work :(
@@Los3Positivos just check the zip file from lshqqytiger, it won't work because it is not a git repository. Therefore, any git command won't work for it. I suggest you get a stable network and use git clone as shown in the tutorial. Cheers 😉
hola. no esta usando la gpu, usa la ram y la cpu, tarda horas en hacer algo, sin mencionar el arranque que tambien tarda. tengo una rx 5700 8gb vram y 16 ram. antes funcionaba bien, y ahora que reinstale haciendo el "downgrade" no. alguna solucion? algun comando que obligue a usal solo la gpu? tampoco me funciono el arg lowvram. gracias!
Good to hear your problem is solved, I believe the problem is specific to different pc build. The setting I show works best for my setup. Have a great day
If you are referring to stable diffusion, it does take a long time to run for the first time. That's why I advise to grab a coffee and do something else. The video is fast-forward so the viewers don't have to wait and watch the installation in real time 😉
Holly Hat Batman... Your video you directed me to worked great!!! Your A GOD! 3 days, 12 videos, 3 different ways to install.. and yours (This one is the only one that worked!!! Thanks for the work! Who ever replied to my call for help! Thanks.. if it was you.. Tong? Then I am buying you a cup of coffee and a donut! :) Simply the best!
@@tongtongrizhi oh-oh.. I was running fine.. I closed it.. to see if it would reload... and it did... but now I only get a black box when I try to generate anything.. big black square a output.. any thoughts?? I am watching it run through the process on my CMD.. it running till 100% done but only Black Square shows.. ? have you seen that before?
@@tongtongrizhi i was doing just that as I guess you were replying.. it worked fine now... Thanks for the quick replies.. Really.. how do i buy you a cup of coffee.. and a donut.. :) love to send you a few Thai Baht.. or $'s :)
Mine is 4-6 s/it for 6600xt. Onnx and olive improve the speed a little but the cons outweigh the speed. Less sampler method to choose and use up a lot of disk space. You can move the stable diffusion folder to SSD, it helps.
If you want to install it without downgrading it, which is by default 1.7.0, all you have to do is skip the git checkout part. The rest is the same and add --use-directml in COMMANDLINE_ARGS at webui-user.bat
Not normal by any means, it could be a few reason: 1. You are using sdxl checkpoint model, don't use sdxl model, they are slow to load and slow to generate image 2. If you are using 1.7.0, you did not add --use-directml to COMMANDLINE_ARGS, please make sure you add set COMMANDLINE_ARGS= --lowvram --disable-nan-check --no-half --precision full --no-half-vae --opt-sub-quad-attention --opt-split-attention-v1 --use-directml in webui-user.bat 3. You generate image way bigger than 768 x 768 pixel, keep the size between 512 and 768. 4. It could be that you are not using the same github repository as the one I did in this tutorial.
@@tongtongrizhi i did whatever you did in this tutorial also I'm usin 1.6 version please make more videos on it bro you are the only one making amd videos in such beautiful detailed way
Sure, do let me know what sort of video you wish to see in the future, I try to make them within my ability. Your gpu is slightly better than mine. It took me around 1 min 30 seconds-ish to generate 512x512 px image and 3 min-ish for generating image with lora and controlnet. If you follow my tut, I assume you are using 1.6.1 version. Maybe you could screenshot your 1. txt2img setting 2. task manager > performance to my email at heinlein_w@hotmail.com so I try to track what wrong with it.
6 หลายเดือนก่อน
AttributeError: module 'onnxruntime' has no attribute 'SessionOptions'
Wait what? This is directml, not onnx. How did you even get this error?👀
6 หลายเดือนก่อน
@@tongtongrizhi I follow all steps and when I use webui user, start to install and appears that error, next I delete the file even folder and works SD for one time, in certain moment, A1111 stops to work, close it and cmd windows too, stats "webui user" again and appears the onnxruntime error again.
@ what file and folder you delete? I think you have to reinstall everything as you watch the video
6 หลายเดือนก่อน
@@tongtongrizhi Yes I do it twice, and watching other videos related. This is thel cmd message venv "H:\SD 2\stable-diffusion-webui-directml\venv\Scripts\Python.exe" Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)] Version: v1.9.0 Commit hash: 174310096385aa33d7b174f231650a941d73f19b Traceback (most recent call last): File "H:\SD 2\stable-diffusion-webui-directml\launch.py", line 48, in main() File "H:\SD 2\stable-diffusion-webui-directml\launch.py", line 39, in main prepare_environment() File "H:\SD 2\stable-diffusion-webui-directml\modules\launch_utils.py", line 658, in prepare_environment from modules.onnx_impl import initialize_olive File "H:\SD 2\stable-diffusion-webui-directml\modules\onnx_impl\__init__.py", line 17, in class DynamicSessionOptions(ort.SessionOptions): AttributeError: module 'onnxruntime' has no attribute 'SessionOptions' Presione una tecla para continuar . . . The folder I delete for works for one time is this: "venv"
6 หลายเดือนก่อน
@@tongtongrizhi In a forum I read someone find the solution and is: "change the direction coz phytom are running in the same folder" I don't understand this. That I found in a forum, but I don't understand how to do that: what is your current working directory? I suspect you are starting python from the onnxruntime source directory? if so, change to any other directory and start python again. if your current working directory has a subdirectory named onnxruntime, python will try to import that subdirectory as a module , which obviously won't work.
Tong: oh-oh.. I was running fine.. I closed it.. to see if it would reload... and it did... but now I only get a black box when I try to generate anything.. big black square a output.. any thoughts?? I am watching it run through the process on my CMD.. it running till 100% done but only Black Square shows.. ? have you seen that before?
Hey Tong.. Sent you a little thank you.. donated so you can grab a cup of Coffee and a Bagel.. Sincerely thank you for the help... I might have another question for you though... tomorrow.. 555 "HaHaHa".. Concerning Local ran GPT Chat uncensored on PC.. I will bother you then if that is okay?!
@@ghosthunters.network you can download this at here gpt4all.io/index.html It's exe file so the installation is straightforward. Then you can download the model you need, uncensor model also included in the download list
This installation is based on Stable Diffusion web UI with DirectML by lshqqytiger
github.com/lshqqytiger/stable-diffusion-webui-directml#stable-diffusion-web-ui-with-directml
Installing Python 3.10.6
www.python.org/downloads/release/python-3106/
Installing Git
git-scm.com/download/win
Git Clone Code
git clone github.com/lshqqytiger/stable-diffusion-webui-directml && cd stable-diffusion-webui-directml && git submodule init && git submodule update
Git Checkout Code
git checkout 03eec1791be011e087985ae93c1f66315d5a250e
Create Virtual Environment
python -m venv venv
venv\Scripts\activate
Checkpoint Model
civitai.com/models/43331/majicmix-realistic
Commandline Args Code
--lowvram --disable-nan-check --no-half --precision full --no-half-vae --opt-sub-quad-attention --opt-split-attention-v1
Donate me at
Kofi: ko-fi.com/tongtong
Paypal: tongtong83@gmail.com
Very good.
Now everything is running without errors, even Upscaling.
Thanks you. Best Tutorial.
I just update to 1.8 and its broke on my 5700 xt setting
you really help bro, its working find on lighting and turbo SDXL
{love you tutorial:1.9}
You are welcome 😉
If you want to improve generation performance, and reduce to 5-7 steps while maintaining quality, I recommend using a Lora model ( lcm-lora-sdv1-5 )
You have to add "Add Network to Prompt": in UI settings you can add SD Lora to appear
thank you thank you thank you, I am now able with my little Ryzen 5 and integrated GPU Vega 11 to use it one hundred per cent, before to do a 512x512 Model 1.5 it took 7 minutes, now with Directml it takes ABOUT 1 MINUTE!!! You are great!!!😇😁🤩
You are welcome. If you plan to switch 1.7.0, I have another video explains how to do so at th-cam.com/video/WQikancF2oE/w-d-xo.html
Thanks again, I will follow your video closely!!!
@@fabiofox5878 thanks so much! Really appreciate it 😊
Try this before anything else:
Uninstall the python versions from your system and install only 3.10.6 (Add to Patch)
This will also ensure that there are no errors during the process. Other versions with the Patch get in the way (at least it works for me)
congrats you got a new sub
😁
Thank you so much! 😊
@@tongtongrizhi bro can you make tutorial on how to add reActor to the extentions . i'm asking you to as you make as the tutorial that you make actually work
actually work.
@@grenadegamerzyt5439 I put this in my next video list
@@tongtongrizhi thanks
thankyou so much man
You are welcome 🤗
Thank you so much i finally can use my amd gpu Thanksss
You are welcome 😉
@@tongtongrizhi is there a way we can use amd gpu for text gen?
@@Mewmew-y4m do you mean txt2img generate? If so, yes you can
@@tongtongrizhi no like oobabooga text-generation-webui for amd gpu
@@Mewmew-y4m that I not sure, it's the first time I ever heard it from you
I had to install SD as a zip file since I kept getting errors, but now I cant get the "Git checkout code" to work. What Im I doing wrong?
Not all stable diffusion sources are the same. Where did you get your zip from? The hash code for git checkout only works specifically on Stable Diffusion web UI with DirectML by lshqqytiger. Any other source other than this won't work. I suggest you follow every steps in the video without skipping any of them. Hope it helps
@@tongtongrizhi I got it directly from your link (by lshqqytiger), could not clone it since I have an unstable network. I had to resort to the Zip file provided in the green code button. But now the "Git checkout code" does not work :(
@@Los3Positivos just check the zip file from lshqqytiger, it won't work because it is not a git repository. Therefore, any git command won't work for it. I suggest you get a stable network and use git clone as shown in the tutorial. Cheers 😉
@@tongtongrizhi Thanks! :)
hola. no esta usando la gpu, usa la ram y la cpu, tarda horas en hacer algo, sin mencionar el arranque que tambien tarda. tengo una rx 5700 8gb vram y 16 ram. antes funcionaba bien, y ahora que reinstale haciendo el "downgrade" no. alguna solucion? algun comando que obligue a usal solo la gpu? tampoco me funciono el arg lowvram. gracias!
Do you mean it doesn't use GPU after you downgrade to 1.6.1? As far as I know, only 1.7.0 need to add --use-directml in COMMANDLINE_ARGS to use GPU
solucionado!!! fue eliminar lo siguiente del webui-use.bat: " --no-half --precision full --no-half-vae --opt-split-attention-v1"
Good to hear your problem is solved, I believe the problem is specific to different pc build. The setting I show works best for my setup. Have a great day
thank you, God bless you@@tongtongrizhi
When i input "venv\Scripts\activate", it says "the system cannot find the path specified".
Have you create the venv environment with "python -m venv venv"
thanks also make a video about how to install Photomake and fooocus with amd gpu
Sure, thanks for the support ☺️
@@tongtongrizhiyes please
ty
Welcome 🤗
Why it seems like, mine is stuck at "Installing requirements"
If you are referring to stable diffusion, it does take a long time to run for the first time. That's why I advise to grab a coffee and do something else. The video is fast-forward so the viewers don't have to wait and watch the installation in real time 😉
@@tongtongrizhi yes yes, it's done, thanks a lot😂☕
@@KinoYukino no prob 👍
I bow to you, it works 6700xt
Welcome, I'm also an AMD GPU user 🤗
@@tongtongrizhi what about rx 580??
@@ms.uglyduck5501 You could give it a try, I think it should work
Holly Hat Batman... Your video you directed me to worked great!!! Your A GOD! 3 days, 12 videos, 3 different ways to install.. and yours (This one is the only one that worked!!! Thanks for the work! Who ever replied to my call for help! Thanks.. if it was you.. Tong? Then I am buying you a cup of coffee and a donut! :) Simply the best!
Glad to know this version works for you
@@tongtongrizhi oh-oh.. I was running fine.. I closed it.. to see if it would reload... and it did... but now I only get a black box when I try to generate anything.. big black square a output.. any thoughts?? I am watching it run through the process on my CMD.. it running till 100% done but only Black Square shows.. ? have you seen that before?
@@ghosthunters.network never had that problem, try restart your PC and launch it again see if the problem persist
@@tongtongrizhi i was doing just that as I guess you were replying.. it worked fine now... Thanks for the quick replies.. Really.. how do i buy you a cup of coffee.. and a donut.. :) love to send you a few Thai Baht.. or $'s :)
@@ghosthunters.network Many thanks, you could do that at here
Donate me at
Kofi: ko-fi.com/tongtong
Paypal: tongtong83@gmail.com
thank you! its worked! but i have bad performance ~1.7 it/s on 7800 XT :(
Mine is 4-6 s/it for 6600xt. Onnx and olive improve the speed a little but the cons outweigh the speed. Less sampler method to choose and use up a lot of disk space. You can move the stable diffusion folder to SSD, it helps.
7800xt works on 11.8it/s only in ubuntu w/ROCm, but you have to accept without Windows, WSL also not working
@@whoopeerobottech638 I'm still waiting for rocm for windows, wonder when it ever get released
Why is my speed at 1.9 it/s on a 7900 XTX.......??
If you want faster, consider using zluda
@@tongtongrizhi Actually i had to change my settings i get 5-8 now but i heard zulda theres alot of models that arent working with it.. so...
@@uxot I only use basic 1.5 model
El directml no se lo banca?
I don't understand the language you write but this version natively supports directml
@@tongtongrizhi Funciona cumpa , lo estoy usando, anda perfecto el Directml
@@tvanime6747 😉 great to hear that
Whats the github code for 1.7.0
If you want to install it without downgrading it, which is by default 1.7.0, all you have to do is skip the git checkout part. The rest is the same and add --use-directml in COMMANDLINE_ARGS at webui-user.bat
@@tongtongrizhi thanks brother ❤️ it takes 11 mins to generate an image and sometimes it gets cancelled ( txt2img) with my amd rx 6650xt is it normal?
Not normal by any means, it could be a few reason:
1. You are using sdxl checkpoint model, don't use sdxl model, they are slow to load and slow to generate image
2. If you are using 1.7.0, you did not add --use-directml to COMMANDLINE_ARGS, please make sure you add set COMMANDLINE_ARGS= --lowvram --disable-nan-check --no-half --precision full --no-half-vae --opt-sub-quad-attention --opt-split-attention-v1 --use-directml in webui-user.bat
3. You generate image way bigger than 768 x 768 pixel, keep the size between 512 and 768.
4. It could be that you are not using the same github repository as the one I did in this tutorial.
@@tongtongrizhi i did whatever you did in this tutorial also I'm usin 1.6 version please make more videos on it bro you are the only one making amd videos in such beautiful detailed way
Sure, do let me know what sort of video you wish to see in the future, I try to make them within my ability. Your gpu is slightly better than mine. It took me around 1 min 30 seconds-ish to generate 512x512 px image and 3 min-ish for generating image with lora and controlnet.
If you follow my tut, I assume you are using 1.6.1 version. Maybe you could screenshot your
1. txt2img setting
2. task manager > performance
to my email at heinlein_w@hotmail.com so I try to track what wrong with it.
AttributeError: module 'onnxruntime' has no attribute 'SessionOptions'
Wait what? This is directml, not onnx. How did you even get this error?👀
@@tongtongrizhi I follow all steps and when I use webui user, start to install and appears that error, next I delete the file even folder and works SD for one time, in certain moment, A1111 stops to work, close it and cmd windows too, stats "webui user" again and appears the onnxruntime error again.
@ what file and folder you delete? I think you have to reinstall everything as you watch the video
@@tongtongrizhi Yes I do it twice, and watching other videos related. This is thel cmd message
venv "H:\SD 2\stable-diffusion-webui-directml\venv\Scripts\Python.exe"
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: v1.9.0
Commit hash: 174310096385aa33d7b174f231650a941d73f19b
Traceback (most recent call last):
File "H:\SD 2\stable-diffusion-webui-directml\launch.py", line 48, in
main()
File "H:\SD 2\stable-diffusion-webui-directml\launch.py", line 39, in main
prepare_environment()
File "H:\SD 2\stable-diffusion-webui-directml\modules\launch_utils.py", line 658, in prepare_environment
from modules.onnx_impl import initialize_olive
File "H:\SD 2\stable-diffusion-webui-directml\modules\onnx_impl\__init__.py", line 17, in
class DynamicSessionOptions(ort.SessionOptions):
AttributeError: module 'onnxruntime' has no attribute 'SessionOptions'
Presione una tecla para continuar . . .
The folder I delete for works for one time is this: "venv"
@@tongtongrizhi In a forum I read someone find the solution and is: "change the direction coz phytom are running in the same folder" I don't understand this.
That I found in a forum, but I don't understand how to do that:
what is your current working directory? I suspect you are starting python from the onnxruntime source directory? if so, change to any other directory and start python again. if your current working directory has a subdirectory named onnxruntime, python will try to import that subdirectory as a module , which obviously won't work.
Tong: oh-oh.. I was running fine.. I closed it.. to see if it would reload... and it did... but now I only get a black box when I try to generate anything.. big black square a output.. any thoughts?? I am watching it run through the process on my CMD.. it running till 100% done but only Black Square shows.. ? have you seen that before?
Hey Tong.. Sent you a little thank you.. donated so you can grab a cup of Coffee and a Bagel.. Sincerely thank you for the help... I might have another question for you though... tomorrow.. 555 "HaHaHa".. Concerning Local ran GPT Chat uncensored on PC.. I will bother you then if that is okay?!
Thanks, I received it already. I used the uncensored chatgpt before but I deleted it due to low disk space, I try find which one I use last time
@@tongtongrizhi thanks.. that could be real helpful.. Much Apprciate.
@@ghosthunters.network you can download this at here
gpt4all.io/index.html
It's exe file so the installation is straightforward. Then you can download the model you need, uncensor model also included in the download list