Nodes/ControlNet Auxiliar/modaux: lineart_anime
ComfyUI Node

modaux: lineart_anime

Anime line art from a noisy screenshot, no clean-up pass required

By madtunebk·Created 2 years ago·Updated 2 years ago· 17
modaux: lineart_anime
  • image
  • IMAGE
detect_resolution512
image_resolution512

There's a version of lineart trained for anime, and it's a different beast. Screenshots of anime and manga are full of compression noise, cel shading, and half-tones - run a general edge detector over that and you get static, not outlines. controlaux_lineart_anime (modaux: lineart_anime) is the variant built to ignore the noise and pull out the actual line work: clean, consistent, cel-style lines that read like the inking pass of a real illustration. If you're feeding anime sources into a ControlNet pipeline - redrawing a character in a new pose, holding a design steady, or just turning a frame into a usable line map - this is the node.

It's part of the ControlNet Auxiliar pack, a thin wrapper over Hugging Face's controlnet_aux. The Annotators repo's lineart_anime model behaves like the "denoise" variant people describe wanting: trained on anime line art so that noisy source images come out clean instead of carrying their compression artifacts through to the condition.

How it works

A single network (netG.pth) downloads from lllyasviel/Annotators on first run, and then the node is one-shot: image in, clean anime line art out. That's the whole mechanism, and it's worth saying plainly because it makes this the simplest node in the pack.

Inputs that matter

There are no optional parameters. Exactly three inputs:

  • image - your anime frame, scan, or render.
  • detect_resolution / image_resolution - both default 512 (range 256–1024). Detection happens at the first, output resizes to the second.

The single IMAGE output is the line map. Pair it with a lineart_anime ControlNet checkpoint - on an anime-capable base like an Illustrious-line model, that combination is a solid, repeatable way to hold a drawing's structure. A modest ControlNet weight, lineart for structure and a reference for identity, and you're running the same character-consistency recipe that made lineart famous, in anime form.

Install

Manager → search "ControlNet Auxiliar", or:

cd ComfyUI/custom_nodes
git clone https://github.com/madtunebk/ComfyUI-ControlnetAux.git
cd ComfyUI-ControlnetAux
pip install -r requirements.txt   # timm, controlnet-aux==0.0.7, mediapipe

Restart, and run it once before you need it so the single weight file is already on disk.

The honest caveat

The whole pack is dormant (last commit mid-2024) and pins controlnet-aux==0.0.7, so don't plan on updates. And while this node is perfectly serviceable, the actively-maintained "ControlNet Auxiliary Preprocessors" pack ships the same anime lineart preprocessor plus a dedicated lineart_anime_denoise mode - if you're building a serious anime pipeline, you'll eventually want that. For a quick, zero-fuss line map from an anime frame, this node does exactly what it says and nothing more.

CategoryControlNet Auxiliar

Inputs (3)

NameTypeDefaultDescription
imageIMAGE
detect_resolutionINT512256–1024
image_resolutionINT512256–1024

Outputs (1)

NameTypeDescription
IMAGEIMAGE