---

name: krita-ai-diffusion description: "Routes Krita AI Diffusion plugin tasks across inference workflows, ComfyUI and cloud resources, Qt workspaces, custom graphs, and Krita document image state." disable-model-invocation: true metadata: disco-role: operating license: GPL 3.0

krita-ai-diffusion

Use this repo skill for Krita AI Diffusion, a Python plugin that adds diffusion image generation, inpainting, live preview, upscaling, animation, custom ComfyUI graph execution, and model/server management to Krita.

This skill is self-contained. Use the bundled references and scripts here; do not send users to source-checkout docs, tests, examples, or scripts as runtime instructions. Treat actual image generation, model downloads, managed ComfyUI installation, and cloud API calls as side-effecting operations that require an explicit user request.

First safe checks

From any checkout or installed package environment, prefer safe read-only checks before launching Krita, ComfyUI, cloud requests, downloads, or generation:

python scripts/check_krita_ai_diffusion_environment.py --static-only
python scripts/list_krita_ai_diffusion_resources.py --summary

If the package is importable and the user needs live API facts:

QT_QPA_PLATFORM=offscreen python scripts/check_krita_ai_diffusion_environment.py --strict
python scripts/list_krita_ai_diffusion_resources.py --parse-url localhost:8188

Known baseline for this generated skill:

• Plugin package version: 1.52.1
• Resource catalog/server version: 1.52.0
• Safe verification scope: CPU/package inspection, static graph inspection, and selected fast tests. GPU generation, managed server install/download, external ComfyUI runtime, and cloud service execution are optional/unselected.

Read references/repo-provenance.md before checking staleness or refreshing this skill. Read references/development-and-testing.md for contributor commands and test-selection policy. Read references/troubleshooting.md for cross-cutting install/import, Qt, server, model, and generation failure recovery.

Route by user task

• Build, inspect, serialize, or debug image generation requests and ComfyUI workflow lowering: use sub-skills/inference-workflows/SKILL.md. Covers WorkflowInput, WorkflowKind.generate, inpaint, refine, upscaling, control image workflows, prompt/style/LoRA preparation, inpaint mode detection, cost/pass estimates, and offline request inspection.

• Connect to, configure, inventory, or troubleshoot ComfyUI/cloud/managed server resources: use sub-skills/server-resources/SKILL.md. Covers ComfyClient, CloudClient, Server, resources.py, required and optional custom nodes, model catalog/resource IDs, URL normalization, backend selection, install/download safety, and server error parsing.

• Work with Qt/Krita model state, workspace routing, jobs, queue behavior, or UI settings: use sub-skills/ui-workspaces/SKILL.md. Covers DocumentModel, Workspace, QueueMode, generation/live/upscale/ custom/animation workspaces, ConnectionState, progress/error kinds, settings persistence, and headless Qt constraints.

• Import, validate, parameterize, or run custom ComfyUI Graph workspace workflows: use sub-skills/custom-graphs/SKILL.md. Covers ETN Krita placeholder nodes, workflow_parameters, graph UI metadata, custom generation modes, live/animation graph behavior, and static custom workflow inspection.

• Manipulate Krita documents, layers, regions, image/mask data, prompt text, styles, metadata, and persistence: use sub-skills/document-image-state/SKILL.md. Covers Image, ImageCollection, Mask, Bounds, Extent, layer tokens, prompt comments/wildcards/LoRA extraction, style JSON, PNG metadata, document persistence, regions, and layer-state troubleshooting.

Package and runtime notes

• The plugin is designed for Krita's embedded Python interpreter. Outside Krita, tests and safe inspection rely on Qt/PyQt and the repo's mock Krita module.
• The source package intentionally vendors ai_diffusion.websockets for plugin releases. If importing ai_diffusion from a source checkout fails with a missing websockets message, initialize or package the vendored websockets submodule before claiming the plugin is installed correctly.
• Avoid third-party runtime assumptions beyond Qt/PyQt and websockets unless a workflow explicitly covers server-side ComfyUI packages.
• api.py/WorkflowInput is the boundary for image generation requests: everything relevant to a generation job must be represented there before it is sent to local ComfyUI or cloud clients.

Verification summary

This skill was verified with static frontmatter/link/path checks, bundled script compilation and help checks, package import/API smoke checks, and selected fast native tests that do not require generation backends. No import into DisCo's managed repo-skill library was performed because the user requested not import.