NovelAI Lineart Processor

Feed it a colored, shaded image and it hands you back the line art underneath - the clean outlines with the rendering stripped away. This is one of NovelAI's Director Tools, called through the API the same way the rest of this pack's utility nodes are. If you've ever used a ControlNet lineart preprocessor to extract structure from a reference image, the goal here is the same: get a clean line drawing out of a finished piece. The difference is this one runs on NAI's own model rather than a local extractor, which means the line quality follows NAI's anime-tuned aesthetic rather than a general-purpose edge detector.

What it's actually for

Line extraction is a workhorse move in this ecosystem, not a novelty. Pull clean lines from a finished illustration and you can recolor it from scratch, use it as ControlNet conditioning to keep a character's structure consistent across a batch of generations, or hand it to a colorist workflow (human or AI) that wants lines without committed shading decisions baked in. Because this is NAI's own tool rather than a local lineart_anime or lineart_realistic preprocessor, expect output tuned toward the same anime-illustration style NAI Diffusion itself specializes in, rather than the more general-purpose behavior you'd get from a local extractor trained on broader data.

The inputs and outputs that matter

No knobs here - one input, one output:

• input_image (IMAGE, required) - the image you want line art extracted from.

Output is a single IMAGE: the extracted line drawing. Wire it straight into a Save Image, or back into the NovelAI generate node's input_image if you want to use the lines as an img2img base for a fresh render.

Installing it

Part of the same pack as every other node 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, then paste your NovelAI API key into the pack's config.py. This node hits NAI's API on every run, so the key has to be valid before it'll do anything - there's no local fallback.

Common issues & troubleshooting

Auth failures first - if this node errors immediately, check config.py for a missing or expired key before assuming anything else is wrong; it's the shared failure mode across this whole pack.

Wrong tool for the job - if what you actually wanted was ControlNet conditioning to guide a new generation's structure, remember this node just gives you an image back, not a control map wired into a ControlNet Apply node. You'd still need to load a ControlNet model and preprocessor pipeline separately if that's your goal; this node is closer to a standalone image-editing step than a conditioning generator.

Style mismatch - because the extraction is done by NAI's own model, don't expect it to behave identically to a general-purpose lineart preprocessor on non-anime-style input. Photographic or heavily painterly source images may extract less cleanly than illustration-style ones, since that's not the domain NAI's tools are tuned for.

Billing - like the rest of this pack, every call spends against your NovelAI subscription. Fine for one-off extraction, worth batching deliberately rather than running loose over a large image set.