NAI 转换草图
Turn any image into a sketch, NovelAI-style
- image
- 图像
- 统计
A small, single-purpose node in a pack full of big machines. NAISketchV2 takes an image and returns its sketch version - the loose, line-heavy translation of the source that sits one step up from a pure line-art extraction. It's one of NovelAI's Director Tools exposed directly, and it costs Anlas per call because NovelAI's model, not your GPU, does the drawing.
What it is
One input, image; two outputs, 图像 (the sketch) and 统计 (a string reporting what came back). No knobs, no sliders, no model picker. That's the whole node. If you want more control you're in the wrong place - this is the "just do it" version of the feature.
The pack ships a whole family of these Director Tools as sibling nodes - line art extraction, colorize, emotion, declutter, background removal - and sketch slots in next to the line-art one. The practical difference between sketch and line art: line art strips the image to clean linework, while the sketch pass keeps more of the original's looseness and shading. Which you want depends on whether you're building an inking stage (line art) or a rough draft (sketch).
What you'd use it for
- Colorization input. The pack's own colorize tool takes a line art or sketch and colors it back in. Run a sketch, tweak it, then colorize for a hand-drawn pipeline entirely inside NovelAI's ecosystem.
- Style exploration. A sketch version of an image reads differently, and if you're trying to match a sketchy art style, feeding a sketch back through generation with low strength can push the output toward it.
- As a first pass. Some people use sketch as an intermediate to check composition before committing to a full paid generation.
Honest caveat: this is a paid API call for something you can approximate locally with edge detection in seconds. The value is NovelAI's aesthetic interpretation - the sketch looks like an anime sketch, not a Canny edge map. If that aesthetic matters, it's worth the Anlas; if you just need linework, a free local node does it.
Install and setup
Part of ComfyUI-NovelAI-GENYTOOLS. Manager search, or:
cd ComfyUI/custom_nodes
git clone https://github.com/XTOGENY/ComfyUI-NovelAI-GENYTOOLS.git
pip install -r ComfyUI-NovelAI-GENYTOOLS/requirements.txt
Restart, set your NovelAI Persistent API Token under NovelAI 2.0: Persistent API Token.
Gotchas
- One API call, one image, one Anlas charge per run - and it re-runs every time the graph executes, so bypass it when you're not iterating on it.
- No parameters means no tuning. If the sketch is too loose or too tight, you're better off with the line-art node or a local preprocessor.
- It only takes a single image in. If you feed it a batch, it handles one at a time under the hood.
- The stats output tells you how many results came back - a sanity check that the Director Tools call succeeded before you inspect the pixels.
Inputs (1)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | — |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| 图像 | IMAGE | — |
| 统计 | STRING | — |