[Inference.Core] Manga Lineart (aka lineart_anime_denoise)

Manga pages and old anime screenshots come with baggage a clean digital illustration doesn't: halftone dot patterns from print scanning, JPEG compression blocking, screentone textures standing in for shading. Point a regular line extractor at that and it dutifully traces every halftone dot as if it were a real line, giving you a speckled, unusable mess. This node - better known by its underlying name, lineart_anime_denoise - is built specifically to clean that noise up before extracting lines, so scanned manga and rough anime screenshots come out as usable line art instead of static.

How it works

It's the same family as this pack's Anime Lineart extractor, but with a denoising step built into the pipeline specifically tuned for the artifacts scanned manga and compressed screenshots carry - halftone patterns, scan noise, screentone shading - rather than assuming clean digital-native source art the way the plain Anime Lineart node does. Practically, that means it's doing two jobs in one pass: cleaning the image, then extracting lines from the cleaned result - which is also why it's a touch heavier than the plain Anime Lineart node on the same hardware.

The inputs and outputs that matter

• image - required, your source manga page or anime screenshot.
• resolution (default 512) - working resolution.

One output: IMAGE, the cleaned-up line art map, wired into an anime-style lineart ControlNet.

How to install it

ComfyUI Manager: search ComfyUI-Inference-Core-Nodes, install, restart. By hand:

cd ComfyUI/custom_nodes
git clone https://github.com/LykosAI/ComfyUI-Inference-Core-Nodes

then run install.py, or pip install -e .[cuda12] (or .[cuda] / .[rocm] / .[directml] / .[cpu]). Restart ComfyUI. The model weights download from Hugging Face the first time you run the node.

Common issues & troubleshooting

If your source is a clean, digitally-drawn illustration rather than a scan or screenshot, you probably don't need this node at all - this pack's plain Anime Lineart node is the simpler, faster choice for source art that's already noise-free, and running the denoising pass on already-clean art can occasionally soften lines you'd rather keep crisp.

If results still look speckled or noisy after running through this node, the source scan may just be low-resolution or heavily compressed to begin with - no denoising pass fully recovers detail that was destroyed by the original compression. In that case, try sourcing a higher-quality scan or screenshot before the preprocessing step rather than expecting this node to fix it downstream. As with the rest of the line-art family, the output only becomes conditioning once it's wired into a matching lineart ControlNet checkpoint - and it's specifically an anime-trained checkpoint you want here, since despite "manga" in the name this is still part of the anime-lineart family rather than the photographic one.