Krita Style

In the Krita AI Diffusion plugin, a "style" is a saved bundle: which checkpoint to use, the sampler and scheduler, step count, guidance, and the baked-in prompt fragments that give a look its character. ETN_KritaStyle is the node that unpacks one of those style presets into a ComfyUI workflow, so the graph downstream is configured entirely by the style the user picked in Krita's UI - no hand-set sampler settings, no hardcoded checkpoint. It's the configuration node in the Krita side of Acly's tooling-nodes pack.

This is the piece that makes the plugin feel like a paint app instead of a node editor. The user chooses a style from a dropdown; this node explodes that choice into every wire a normal ComfyUI generation needs. It's the reason someone can run sophisticated Stable Diffusion workflows from inside Krita without ever learning what a scheduler is - one of the big reasons the plugin is a favorite for people who paint.

How it works

You give it a style name and a sampler preset, and it outputs the full kit: a loaded model, its CLIP and VAE, the positive and negative prompt strings the style defines, and the sampler settings (sampler name, scheduler, steps, guidance). The plugin manages the actual style definitions; this node surfaces the selected one into the graph as ordinary ComfyUI outputs that fan out to your loaders, CLIP Text Encodes, and KSampler.

Notice the shape: it emits the prompts as strings, not pre-encoded conditioning, and the sampler settings as discrete values. That's deliberate - it means the workflow can combine the style's prompt with a user's per-generation prompt, or override a setting, before encoding. The style is a starting configuration, not a locked pipeline.

The inputs and outputs that matter

• name (STRING, default Style) - which style preset to load. The plugin sets this to match the user's selection.
• sampler_preset (COMBO) - the sampler preset to resolve into concrete sampler/scheduler/steps/guidance values.

Outputs - essentially a whole generation's worth of configuration:

• model, clip, vae - the loaded checkpoint pieces. Wire like any checkpoint loader's outputs.
• positive prompt / negative prompt (STRING) - the style's prompt fragments, to combine with the user prompt and then run through CLIP Text Encode.
• sampler name, scheduler (COMBO), steps (INT), guidance (FLOAT) - the sampler config, straight into your KSampler (or KSampler Advanced).

How to install it

Part of the tooling-nodes pack. ComfyUI Manager: search ComfyUI Nodes for External Tooling, install, restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/Acly/comfyui-tooling-nodes.git

then restart. To use styles as designed, install the Krita AI Diffusion plugin, which defines and manages the presets this node reads. Whatever checkpoint the chosen style names still has to be present in your ComfyUI models folder.

Common issues & troubleshooting

Style won't load / model missing. The style references a checkpoint by name; if that file isn't in your ComfyUI models/checkpoints, the load fails. Make sure the style's model is actually installed on the ComfyUI server.

Prompts aren't taking effect. The prompt outputs are plain strings - they still need to go through CLIP Text Encode to become conditioning. If you wired the string straight into the sampler, that won't work; encode it first (usually concatenated with the user's prompt).

Building this by hand? You generally shouldn't. This node is populated and wired by the Krita plugin's generated workflows. Hand-authoring around it is an advanced customization path, not the normal way to use the plugin.