NAI 自动上色

NAI 自动上色 (Colorize) is the colorist of the Director Tools family: you feed it a lineart or sketch and NovelAI paints it. It's the natural partner to the lineart node - extract the lineart from a finished piece, or draw a rough sketch, then let this node colorize it with the palette and style you hint at in a prompt. A hand-drawn doodle becoming a finished colored illustration in one API call is the demo, and it genuinely works.

Inputs:

• image - the lineart or sketch, as an IMAGE.
• prompt - the hint text. This is where you steer the palette and mood: "sunset lighting, warm tones, red kimono" does a lot more than an empty box. Empty is allowed, but colorization without a prompt is a roll of the dice.
• defry - 0–5, default 0. NovelAI's "defry" parameter - how much the result fights the muddy/fried look that low-detail inputs can produce. Higher values push toward cleaner, more defined color; 0 is the literal default, and you'll usually want at least a couple of notches if the source is a rough sketch.

Outputs are 图像 (the colorized IMAGE) and 统计 (result count string). Standard pack wiring: image out → preview or saver, or further editing.

The workflow that makes sense

Colorize really shines at the two ends of a comic/manga pipeline. From a finished black-and-white page you want inked color: feed lineart, prompt the palette, done. From a rough idea you want a color comp: sketch → colorize → iterate on the prompt until the palette lands. It's one-shot, though - there's no strength dial, so the way to adjust is to change the prompt or the source, not to tune a slider. If the result is close but the palette's wrong, re-run with a more specific prompt before you try to fix it in post.

The honest caveats: it's a paid API call (Anlas), it works best on clean lineart - a messy scan with grey noise will produce a messy colorization, so run lineart extraction first if your source is a photo - and "defry" is the one knob you'll actually touch. And like every Director Tool, it only fires when you execute the node; wiring it in is free.

Install

Search ComfyUI-NovelAI-GENYTOOLS in ComfyUI Manager, or git clone https://github.com/XTOGENY/ComfyUI-NovelAI-GENYTOOLS.git into custom_nodes + pip install -r requirements.txt, Persistent API Token in settings.