Common Issues

This document collects some common issues and potential solutions.

Note

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!

Collecting Diagnostics

To make it easier to provide relevant information when reporting an issue, the plugin has built-in diagnostics. It only collects information locally, it does not send it anywhere! You can find it in the plugin settings under Plugin ‣ Collect Diagnostics.

Clicking the button will copy the report to the clipboard, so you can paste it into your GitHub issue. It also opens the contents in a separte window for review. Private information is anonymized as much as possible.

Log files

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.

ComfyUI Workflow export

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.

Installation

Issues related to installation and update of the plugin and its managed server.

Plugin is grayed out in Python Plugins Manager

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!

How do I update to a new version of the Plugin?

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.

Caution

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)

Where are plugin and user files located?

The plugin stores data in up to three locations:

1. <appdata>/krita/pykrita/ai_diffusion - This is the folder where you installed.
2. <appdata>/krita/ai_diffusion - This folder contains user data (settings, styles, logs…)
3. 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/...

Where are files cached? Why is space on C: drive required?

Even when installing the plugin and server on a different drive, significant space is used on the system drive (C:) during installation for temporary files and caches. If you are running out of space, you can control the locations of these files by setting environment variables:

Cache	Default Location (Windows)	Purpose	Environment Variable
HuggingFace	%USERPROFILE%\.cache\huggingface	Model cache	HF_HOME
uv	%LOCALAPPDATA%\uv\cache	Python packages	UV_CACHE_DIR
pip	%LOCALAPPDATA%\pip\cache	Python packages	PIP_CACHE_DIR

It is okay to delete the contents of these folders to free up space. Important files will be re-downloaded as needed.

How do I relocate my server installation?

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.

How do I uninstall the Plugin?

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.

How do I perform a clean re-install of the Plugin?

Follow the installation instructions, it will delete and replace the previous version.

If you want to re-install the server, you can do this from the Connection settings. Choose Manage > Re-install. This will keep your models and settings, but re-download the server software and update it to the latest version.

How to share models from another folder with Krita AI Diffusion plugin?

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:

1. Go to the folder where you installed the server (“Server Path”)
2. Go into the ComfyUI folder
3. Open the file extra_model_paths.yaml in a text editor and change the paths to point to the locations where your models are stored

How do I pass additional arguments to ComfyUI?

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:

1. Locate settings.json in the user data folder (<appdata>/krita/ai_diffusion)
2. Open it with a text editor (notepad will do)
3. 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.

Error while deserializing header

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.

If you are using a managed server installation, you can do this by going to the plugin’s connection settings and clicking Manage > Verify. This will automatically check all models and re-download them if necessary.

Operating System and Hardware

Supported platforms

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.

Note

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”.

Related issues

Image generation is (really) slow

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.

Windows: OSError / WinError: The specified module could not be found

If you encounter errors like this one:

Error loading "C:\User\...\python\Lib\site-packages\torch\lib\c10.dll" 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

Windows: [Errno 2] No such file or directory

If you encounter an error like this:

Server execution error: [Errno 2] No such file or directory:
'C:\...\AppData\Roaming\krita\ai_diffusion\server\ComfyUI\custom_nodes\comfyui_controlnet_aux\ckpts\...[somefile].incomplete'

when trying to use control layers, it is likely that the file path is too long.

Some Windows versions have a limit of 260 characters for file paths. To fix this issue:

• Move the server installation to a shorter path, eg. C:\KritaAI
• Enable long paths in Windows 10/11

macOS: xcrun: error: invalid active developer path (missing xcrun)

Install Xcode command line tools, eg. by running the following command in a terminal:

xcode-select --install

macOS: No images received from remote ComfyUI server

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.

Custom ComfyUI

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.

Note

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.

Related issues

Error: The following ComfyUI custom nodes are missing

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!

If nodes are still reported missing after an update, make sure the node packs are the latest version. A new version of the plugin may rely on nodes added in the most recent versions of the node packs, and cannot function without them.

Sometimes updates via ComfyUI Manager are reported as successful, but the files are not actually updated. Try to delete and reinstall the node pack instead.

Keep in mind ComfyUI has to be restarted after installing new nodes!

Sometimes while troubleshooting you can end up with multiple instances of the ComfyUI server running in the background. This can be difficult to detect and lead to confusion - you may see the nodes working in the web UI, yet the plugin reports them as missing. Check your running processes for leftover python processes, or just log off / reboot to be sure.

Error: Could not find Upscale/CLIPVision/IP-Adapter model

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.

Missing common models: MAT Inpaint (MAT_Places512_G_fp16.safetensors)

See model list for the download link. The most common causes for not finding the model are:

• The model is not in the correct folder. The most direct way is to open the location where you installed ComfyUI, navigate to the models folder, and create a new folder inpaint there (if it doesn’t already exist). Place the model inside.
• You can use extra_model_paths.yaml, but the inpaint folder has to be added there explicitly! Also check for possible typos.
• If you are using Stability Matrix, keep in mind it does its own folder forwarding.
• There may be some other custom node which overwrites the folder paths in a destructive way. Consider disabling unnecessary custom nodes to check if this is the issue.

This is a SD 1.5/XL checkpoint, but the SD 1.5/XL workload has not been installed

Workloads refer to the files listed as Required models. 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.

Error: Hyper LoRA model not found

Make sure you downloaded required LoRA for your base model. 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.

Connecting takes a long time or is stuck

If you see a long delay when connecting, ie.

• “Connecting” is shown for a long time
• “Discovering models” takes a long time this is likely related to detecting available models. Usually model detectiong is a quick process, but in some cases it can take very long. Possible reasons:

1. You have a very large number of checkpoints or diffusion models
2. Your model folder is on a slow or remote drive
3. A virus scanner like Windows Defender is scanning model files

To resolve this, try to reduce the number of models by archieving the ones you don’t actively use, and make sure they are on a fast drive (SSD). Windows users can also try to exclude the model folder from virus scanning. In Windows settings search for “Virus & threat protection” > “Manage settings” > “Exclusions” > “Add or remove exclusions”.