Common Issues
This document collects some common issues and potential solutions.
Important
Before you report a problem on GitHub:
- make sure it cannot be resolved by one of the solutions proposed here
- use the search to find existing issues related to your problem
- gather relevant information such as OS, Krita and Plugin versions, GPU, Log files (see below)
You can create a new Issue here, or ask questions in Discussions or on Discord.
Please note that official Krita channels are not the right place to seek help with issues related to this extension!
If you encounter an issue, it can help to check the log files. You can locate the folder with the "View log files" link in the plugin's connection settings. Please attach logs when you open an Issue on GitHub!
Sometimes it can be helpful to inspect the workflow generated by the plugin in Comfy's web UI. You can enable workflow export in the Interface settings. The exported workflow.json of the last image generation can be found in the same folder as the logs.
The most common reason is that a source package was used for installation (eg using GitHub's Code ‣ Download ZIP). It is automatically generated by GitHub and does not contain everything. Please try with a release package.
Krita will show an error message when you hover over the grayed out entry in the plugin manager. See below for an example. Please make sure to include the error message if you report an issue!
To upgrade a previous version of the plugin, simply follow the instructions for initial installation. The old version will be replaced.
Krita has to be restarted for the update to take effect. If your server installation is managed by the plugin, it may prompt you to do an upgrade. This will automatically reinstall the server software, but without downloading models again.
Warning
Import Python Plugin from File... deletes the previous version and replaces it! If you originally installed a version before 1.15.0, your server installation and settings may be located in the plugin folder. In that case either
- Extract the plugin manually into the installation folder (
.../pykrita/ai_diffusion). Overwrite existing files when prompted. - Or relocate your server installation first (you can move the folder and update the path in the UI)
The plugin stores data in up to three locations:
-
<appdata>/krita/pykrita/ai_diffusion- This is the folder where you installed. -
<appdata>/krita/ai_diffusion- This folder contains user data (settings, styles, logs...) - The server location if you installed it via the plugin. By default it is in the same place as user data (2.) unless you changed it manually.
The full path depends on OS and how Krita was installed. You can find it via the "View log files" link in the plugin's connection settings. Typical paths are:
- Windows:
C:\User\<your-name>\AppData\Roaming\krita\... - Linux:
~/.local/share/krita/... - MacOS:
~/Library/Application Support/krita/...
By default the installer will place the ComfyUI server in <appdata>/krita/ai_diffusion/server. The path can be chosen freely in the UI and point to any other folder! If you already have an installation it's also possible to simply move the folder to a different location, and then change the path in the UI afterwards.
You can remove the plugin simply by deleting the folder where you installed it.
If you installed the server via the plugin, find it at the path you chose for installation. Delete that folder if you have no further use for it.
See plugin file locations for default locations.
Version 1.15.0 and later
Follow the installation instructions, it will delete and replace the previous version.
If you want to delete user files (settings, styles, etc.) see plugin file locations.
Prior to version 1.15.0
If you want to keep settings, styles, or the server installation, delete everything inside the ai_diffusion folder except:
.logs.serverstylessettings.json
On some platforms folders starting with . might be hidden - enable showing invisible files in your file browser to see them.
If you used the plugin to install and set up ComfyUI, but already have Stable Diffusion models in a different location, it is possible to share them:
- Go to the folder where you installed the server ("Server Path")
- Go into the
ComfyUIfolder - Open the file
extra_model_paths.yamlin a text editor and change the paths to point to the locations where your models are stored
Some issues with certain hardware can be solved by passing additional arguments to the ComfyUI server. If you used the plugin installer, it will start/stop the server for you. To add arguments:
- Locate
settings.jsonin the user data folder (<appdata>/krita/ai_diffusion) - Open it with a text editor (notepad will do)
- Find the line with
"server_arguments"and edit the value
For example if you want to add --force-fp16 and --dont-upcast-attention arguments it should look like this:
"server_arguments": "--dont-upcast-attention --force-fp16"
Error during server startup: Could not bind on address (only one usage of each socket address is normally permitted)
If you get this error it usually means another process is already running and using the same port. This could be:
- Another instance of ComfyUI you are running manually
- A server process previously started by the Krita plugin that wasn't shut down correctly
- Some other process on your system which uses the same port (8188)
- An issues with system or firewall settings (unlikely)
The easiest solution is to shutdown the other process which is using the same port. You can do this manually, or just log off / reboot.
If you don't want to shutdown the other process or can't find it, you can also change the port the server runs on. See above on how to change server arguments. Set them to eg.:
"server_arguments": "--port 25608"
You can choose any port number you like.
If you see an error of the kind HeaderTooLarge or MetadataIncompleteBuffer it is most likely because one of the model files is corrupt (has not been downloaded properly). Try to delete and re-download them in order to fix the issue.
To find the model files, open the folder where you installed the server. Model files are located in the models sub-folders. Deleting the file will allow you to redownload them via the installer, or you can do so manually (see the list here).
Krita and the AI Diffusion plugin itself work on Windows, Linux and MacOS with all common hardware. But to run diffusion locally a powerful GPU and sufficient CPU/RAM are also required. The plugin relies on ComfyUI, which supports a growing number of platforms. Unfortunately not all platforms are equally well supported.
NVIDIA GPUs have by far the highest investment, and are therefore the most reliabe. All other hardware (AMD, Intel, Apple) often does not perform as well or occasionally breaks. Support relies on many other projects, often maintained by volunteers, and also the hardware vendors themselves.
Important
Krita AI Diffusion plugin is continually tested on Windows and Linux with NVIDIA GPUs. Other operating systems and hardware may be supported, but their support relies on contributions from users of these platforms.
If you encounter server errors reported by ComfyUI on such systems we cannot always reproduce them, and you will likely find more help by eg. searching the internet for "ComfyUI + your os/gpu".
If you consider generation too slow, here are a few things to check:
- What resolution is your canvas? Start with a low resolution (eg. 512x512) and go up from there.
- Are you using SD 1.5 or SD XL? Note that SD XL uses a minimum resolution of 1024 and requires more resources.
- Are you running out of VRAM? There are tools to track GPU memory usage. If it hits your limit, you may not get an error, but extremely long generation times instead.
- What is your hardware? What is the expected performance for Stable Diffusion? You can find comparisons for many GPUs online.
- What kind of performance do you get when using ComfyUI directly?
- With Krita and the plugin running, go to http://127.0.0.1:8188 in your browser and "Queue Prompt".
- If performance is bad there, the problem is not related to the Krita plugin.
If you encounter errors like this one:
Error loading "C:\User\...\python\Lib\site-packages\torch\lib\c10.dill" or one of its dependencies.
when starting the server, you are probably missing some required system DLL. You can download them from Microsoft.
More information: https://learn.microsoft.com/en-us/cpp/windows/latest-supported-vc-redist?view=msvc-170#latest-microsoft-visual-c-redistributable-version
Install Xcode command line tools, eg. by running the following command in a terminal:
xcode-select --install
If you are connecting to a remote server like a different machine you own, or a cloud service like runpod.ai, and run into the situation where images appear to be generated but don't show up: this might be related to SSL certificates on macOS. See this issue for solutions.
If you choose to manage your own ComfyUI installation rather than using the built-in installer, keep in mind that you are responsible for getting it running, installing the plugin's prerequisites as needed, and keeping it up to date. ComfyUI is a fast-moving project with a huge number of extensions and potential for customization. As a result you may also sometimes encounter problems and conflicts that need to be resolved.
Important
Please understand that we cannot give tech support for individual ComfyUI installations out there. Don't open an issue here if the problem is not related to the Krita plugin.
A lot of related issues can be resolved by carefully following the requirements and its troubleshooting section.
Make sure you installed the nodes listed in the requirements. In case a custom node package is reported to be missing even though you installed it, please check that ComfyUI startup console output lists them as loaded, and does not report errors. Also make sure the nodes can be created in Comfy's web UI!
Make sure you placed the models listed in the requirements in the specified folders. If models are still reported missing, first check if ComfyUI can find them. Open its web UI and create the respective loader nodes ("Load Upscale Model", "Load CLIP Vision", "Load IPAdapter Model"). If they are not listed there they are in the wrong folder or there is an issue with your search paths.
Workloads refer to the files listed in the requirements in the section "Required models ‣ SD 1.5" (respectively SD XL). Because you can use the plugin with either SD 1.5 or SD XL, it is not an error if one of them is missing - but you can only use those checkpoints for which you have all the prerequisites.
Make sure you downloaded and renamed the file as described in requirements. Create a "Load Lora" node in Comfy's web UI and check if the file is listed there. If it cannot find it it's in the wrong folder or there is an issue with your search paths.