Can't get a connection from Krita to local installed Stable Difusion 1111

[FritzHugo3]

Hi there,
I can’t connect Krita with my local installed Stable Diffusion and I have absolutely no idea why.

I have checked following:
Install
ControlNet preprocessors
IP-Adapter
Ultimate SD Upscale
Inpaint nodes
External tooling nodes

Have tested with deactivated firewall

(have not yet installed the needed models – I think its not needed for the Server connection handshake or)

But Krita say always – Custom Server (local or remote):
Error: Network error (Error transferring http:https://my-custom-IP-what-I’m-copy-from-the-BrowserUI/system_stats - server replied: Not Found (Log are ok but only no connection)

And sure – my local WebUI 1111 work absolute fine itself, it is running and Stable Diffusion can make pictures over the Browser

But what are “MyIP/system_stats”? If I type it in the Browser – I get JSON/Rawdata/Headsigne and
Detail: “not found. Are here the problem?

Win11, version: v1.9.3-amd-25-g73a4e8c0  •  python: 3.10.6  •  torch: 2.0.0+cpu  •  xformers: N/A  •  gradio: 3.41.2  •  checkpoint: ef76aa2332

Thx for your time and help.

[ricewind100]

automatic111 not work,
use it
https://github.com/comfyanonymous/ComfyUI

[FritzHugo3]

A ok thx. i havent read that 1111 dont work.
Yes i have now used the complete automated installation for a paralell Krita-Stable Difussion server - little bit much gb both but anyway-this work.

Is it normal that we haven't all the huge option from the webUI in Krita? Or have i dont found it yet?
And no console to see what process work...
Krita now use much to much details and options for my grafikcard in webUI i can handle all by scaling the slider, can say give me multiple picture versions... Is ther an good actual Tutorial video for it? What i have found in YT saw different from the UI

And thx for your answer.

Edit, have tested a while now. Create pictures with normal stable defussion is not a problem but the Krita version i get only one error afer other - AMD 8GB graphic Ram are not enought maybe and i cant find any options to say make lower quality/steps/...

[Drakosha405]

in my opinion, the first thing you should determine for yourself is that this add-on is designed for drawing and not for generating images with a large number of changeable variables. If you want to find the right prompt or any other settings for the desired effect, go to any Web UI where it's easier and faster to do it. and use Krita to finish drawing the details (by adjusting the necessary settings for this add-on). or as the workflow is presented in the video, where you draw images yourself from the very beginning and refine it with the help of AI

if you want to see the process in the console, run comfyui separately (option 2 to connect to the server)

krita does not load the graphics card, and comfyui has the highest performance even on weak graphics cards (I have 1060 6gb). The time increase depends on the number of selected options such as transfer styles or canny, the strength of cfg and the number of steps and some more items. pay attention to Hyper-SD

do you use DirectML with AMD graphic card?

[FritzHugo3]

I can't paint live or expand outpainting, get instand error message for low graphic ram - I want use krita for chance things from pictures who i make before in 1111 - but it is not working here - i don't find the option to say how many steps, how quality or something like this. In 1111 i can say use not

I write single word in Promt house and make the first line = Server execution error: Could not allocate tensor with 268435456 bytes ther is not enough GPU Video Memory avaible - for me no "a large number of changeable variables"

If i have a complete fresh new paper and click Live - it will start to work for 10 seconds befor i write oder draw anything. It is ea fresh installation of krita and server. So i have realy no idea whats wrong.

Do you use DirectML with AMD graphic card? - Yes
AMD Radeon RX 6650XT 8GB
And its the "Let Krita plugin install and run a local server on your mashine with default settings.

Why he say always "268435456 bytes"? who say krita ther are 256GB RAM?
I testet it again and i use a fresh 128x128 (really smal to have not a big calculating) and without anything - I click the Live Button and get the error to low RAM - I mean this is not normal or?

[Drakosha405]

I can't paint live or expand outpainting, get instand error message for low graphic ram - I want use krita for chance things from pictures who i make before in 1111 - but it is not working here - i don't find the option to say how many steps, how quality or something like this. In 1111 i can say use not

staps, cfg to be configured in the style settings, I advise you to read the wiki

And its the "Let Krita plugin install and run a local server on your mashine with default settings.

if self.backend is ServerBackend.directml: await _execute_process("PyTorch", self._pip_install("torch-directml"), self.path, cb)

this line shows that we should have Direct ML selected before installing all the components, did you select CUDA at the first launch

if this is the case, then you need to additionally install torch-directml
pip install torch-directml

[FritzHugo3]

No of couse i have never install CUDA, it makes no sense, its for NVidia, i have an AMD grafigcard. It is a fresh, clean "use DirecML(GPU with autamted Krita plugin install on local mashine (first option. Default i have nothing deactivate or activate on other compounds.

"if self.backend is ServerBackend.directml: await _execute_process("PyTorch", self._pip_install("torch-directml"), self.path, cb)"
i dont now what you want to say to me - whatr is with this text? Never see this or shpold i copy this anywhere?
Also this "if this is the case, then you need to additionally install torch-directml" pip install torch-directml"

128x128

[Drakosha405]

No of couse i have never install CUDA, it makes no sense, its for NVidia, i have an AMD grafigcard. It is a fresh, clean "use DirecML(GPU with autamted Krita plugin install on local mashine (first option. Default i have nothing deactivate or activate on other compounds.

Did you choose "Use DirectML" after installing all the components? if so, the plugin has installed Cuda files.
this option must be selected before the installation begins of comfyui