Krita Style & Prompt

Krita Style & Prompt is the big convenience node of the pack's Krita family - a one-node unpacker that takes a style preset and spits out every ingredient a sampler needs. In Krita AI Diffusion, a "style" bundles a checkpoint, its LoRAs, prompt boilerplate, and sampler settings into a single named preset you pick from a dropdown. This node is what explodes that preset back into discrete wires inside the ComfyUI graph, so the plugin doesn't have to rebuild a dozen loader and encoder nodes every time you switch styles. It's the seam between "the user picked a style" and "the workflow has a model, a CLIP, a VAE, and numbers to sample with."

The wider context is the same as the rest of the krita category: Acly builds both this pack and Krita AI Diffusion, a well-liked way to run Stable Diffusion inside Krita by driving a ComfyUI backend. Styles are how that plugin keeps the UI simple for people who'd rather paint than wire graphs - you configure a style once, then just choose it. Under the hood, though, ComfyUI still needs the model loaded, the LoRAs applied, the prompt encoded, the sampler configured. Rather than emit that whole subgraph, the plugin drops in this single node and lets it resolve the preset at runtime.

The input is deceptively small:

• sampler_preset (COMBO) - the style/preset to unpack. One dropdown, and everything else follows from it.

What comes out is the whole toolkit, which is why the node earns its keep:

• model (with loras) (MODEL) - the checkpoint with the style's LoRAs already applied.
• clip (CLIP) and vae (VAE) - the matching text encoder and VAE.
• positive prompt (evaluated) and negative prompt (evaluated) (STRING) - the style's prompt templates, resolved.
• sampler name (COMBO), scheduler (COMBO), steps (INT), and guidance (FLOAT) - the sampler settings the preset carries.

Wire those into a KSampler (Advanced) or the equivalent and you've reconstructed a full generation pipeline from one node. The "with loras" and "evaluated" labels are the tell that this node did real work - it applied the LoRAs and resolved the prompt templates for you.

Installation is the pack standard: "ComfyUI Nodes for External Tooling" in ComfyUI Manager, or cd custom_nodes && git clone https://github.com/Acly/comfyui-tooling-nodes.git from your ComfyUI directory, then restart. No extra dependencies. But the honest prerequisite is the same as every krita node here - the style presets this reads live in Krita AI Diffusion's configuration, and the plugin installs and manages this pack itself. Outside that plugin, the sampler_preset dropdown has nothing meaningful to choose from.

Which is exactly the gotcha. This is the least portable node in the whole pack. Krita Image Layer at least receives generic image data; this one is tightly bound to the plugin's style system, so dropping it into a standalone browser graph gives you a dropdown pointing at presets that don't exist there. It genuinely only makes sense as part of a plugin-generated workflow. If you're hand-building in ComfyUI and want the same result, you'd assemble it the normal way - Load Checkpoint, LoRA loaders, CLIP Text Encode, and your sampler settings as explicit nodes. That's more boxes on the canvas, but it's the path that actually works outside Krita.

If you're debugging on the plugin side and a generation comes out wrong after a style change, this node is the first place to look - everything downstream inherits the model, prompts, and sampler numbers it emits, so a misconfigured style shows up here before it shows up in the image.