NovelAI Sketch Processor

This one runs in the opposite direction from what you'd expect. Instead of sketch-to-finished-art, it takes a finished (or mostly finished) image and converts it down into something that looks like a rough sketch - loose lines, less commitment to detail. It's another of NovelAI's Director Tools, the same family the Lineart and Declutter processors in this pack belong to, wrapped as its own node.

Why you'd want that

The obvious use is building an img2img base with room to move. A fully rendered image locked into an img2img pass tends to hold its composition and detail too rigidly, even at moderate denoise - you're fighting the finished piece instead of reinterpreting it. Knock it back to a sketch first, and a subsequent generation has more freedom to reinterpret color, lighting, and finish while still roughly following the original layout. It's also just useful on its own if you want a sketch-style variant of an existing image, or a loose base to hand off to manual linework.

How it works

Same shape as the rest of this pack: it's an API call, not local processing. Your image goes to NovelAI, their sketch-conversion tool runs against it, and you get the result back as a normal ComfyUI IMAGE. Nothing runs on your GPU for this step - the only requirement is a valid API key and an active NovelAI subscription behind it.

The inputs and outputs that matter

One input, one output, no tunable parameters exposed:

• input_image (IMAGE, required) - the image you want converted to a sketch.

Output is a single IMAGE - the sketch result. The natural next step is wiring it into the NovelAI generate node's input_image with a moderate-to-high denoise, letting NAI Diffusion reinterpret it from there. It works just as well as a plain output if a loose sketch is what you actually wanted.

Installing it

Same pack, same setup as every other node covered here. ComfyUI Manager: search comfyui_nai_api, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/Creeper-MZ/comfyui_nai_api

Restart ComfyUI, then set your API key in the pack's config.py. This node calls NovelAI's API directly, so it needs a working key before it'll produce anything - there's nothing to configure beyond that one file.

Common issues & troubleshooting

Auth errors first, as always with this pack - a missing or expired key in config.py is the most likely reason this fails outright.

No control over how rough the sketch comes out - since the only input is the image itself, you can't dial in more or less simplification. If the result is too loose or not loose enough for what you need, there's no local parameter to adjust; your options are pre-processing the input differently or accepting NAI's default behavior.

Don't confuse it with Declutter - Declutter cleans up noise while keeping detail roughly intact; this node deliberately throws detail away to produce something sketch-like. If you wanted a tidied-up version of your image rather than a simplified one, use the Declutter Preprocessor node instead.

Billing - another API-metered step. Fine for a handful of conversions, but factor the cost in before looping it over a big batch of source images.