Standard Lineart

Standard Lineart is the plain, dependable line extractor - an algorithmic method that turns any image into clean outlines without loading a neural model. It's fast, it's lightweight, and it's a solid default when you just want lines to feed a lineart ControlNet and don't need the specialization of the anime, manga, or realistic variants. Think of it as the "just give me sensible lines" option: it doesn't have opinions about whether your source is a photo or a drawing, it just extracts.

How it works

Unlike the realistic and anime lineart nodes, which run trained networks, Standard Lineart is a classical image-processing pipeline - a Gaussian blur followed by an intensity-difference step that isolates line-like structures. That's why it needs no model download and runs instantly. It's the same lineart_standard method that pairs with the standard lineart ControlNet, and it's the neutral baseline the other lineart nodes are variations on. For a lot of everyday structure work it's genuinely all you need.

The inputs that matter

Two tuning knobs, and they're the reason to use the standalone node instead of grabbing this through AIO:

• guassian_sigma (default 6 - yes, the parameter name is misspelled in the node, that's cosmetic) - the blur strength before line extraction. Higher sigma smooths the image more, producing fewer, cleaner, bolder lines; lower sigma keeps finer detail but more noise. This is your "detail vs cleanliness" dial.
• intensity_threshold (default 8) - how strong a difference counts as a line. Raise it to keep only the most pronounced outlines; lower it to catch subtler lines (and more noise). Between these two, you can tune from a sparse, bold outline to a dense, detailed one.
• resolution (default 512) - working size; match your render.

The single output is an IMAGE - white lines on black - that feeds a ControlNet Apply node with a lineart ControlNet (control_v11p_sd15_lineart). The node makes the hint; the ControlNet model is separate.

Installing it

ComfyUI Manager: search ComfyUI's ControlNet Auxiliary Preprocessors, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/Fannovel16/comfyui_controlnet_aux
pip install -r requirements.txt

Restart ComfyUI. Because it's a pure algorithm with no weights, this node works immediately - no first-run download, no HuggingFace fetch.

Where people get burned

First, the everyone-does-it one: the lines are inert without a lineart ControlNet loaded and wired to this output. Second, don't sleep on guassian_sigma and intensity_threshold - the defaults are reasonable but not magic, and if your line map is too busy (raise sigma, raise threshold) or too sparse (lower them), those two dials fix it. This is exactly the tuning AIO hides, which is why you reach for the standalone node once you're past exploring. Third, pick the right lineart node for the job: Standard is the neutral all-rounder, but for clean anime art the Anime Lineart node is better, for noisy manga scans the Manga Lineart node denoises first, and for maximum detail AnyLine wins. Standard's virtue is that it's fast, dependency-free, and never wrong-footed by content type - a great default and a great fallback when a fancier extractor is misbehaving.