NAI 移除背景

Background removal is the most commoditized operation in the whole image ecosystem - BiRefNet, rembg, InSPyReNet, a dozen free local nodes. So why would you call NovelAI's paid version? Because it's the one removal tool that knows NovelAI's art style. Local models are great at cutting out subjects, but they don't know what an anime background is supposed to look like when it's gone. This node runs NovelAI's own Director Tools background-removal, and the interesting part is what it gives you back.

What it is

One input, an IMAGE, and three image outputs plus a stats string:

• 遮罩结果 (mask result) - the transparency mask itself, what you'd use to do your own compositing.
• 生成结果 (generated result) - the background-removed image with whatever NovelAI's model filled in behind the subject.
• 混合结果 (blend result) - a blended version of the two, usually the one you actually want to save.

It runs NovelAI's Director Tools "bg-removal" endpoint, which means it costs Anlas per call, unlike the free local nodes. The triple output is the feature: the mask lets you keep full control downstream, the generated version is NovelAI's own take on what should replace the background, and the blend is the compromise that often reads best.

When it beats the local options

The KB's background-removal essay is blunt that the local models are free, fast, and plural - and that the quality frontier moved to hair, fur and semi-transparency, where nobody has fully won. So this node wins on a different axis: style coherence. A BiRefNet cutout gives you a subject and a hard edge; it doesn't understand that an anime illustration's background has a painted quality that a plain transparency doesn't preserve. NovelAI's model keeps the illustration's visual language while removing the scene.

It loses on cost, speed, and batch friendliness. If you're removing backgrounds from a hundred images for a dataset, the free local nodes are the right tool and this isn't. If you're producing one polished illustration and want NovelAI's aesthetic to survive the edit, this is the one that matches.

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

• It's a paid Director Tools call - one request, Anlas charged. The stats string reports what came back.
• Three outputs, one often enough: start by looking at the blend result; go to the mask when you want to do your own compositing (the pack's tiled or inpainting nodes downstream can use it).
• It always re-runs, so re-executing the graph re-bills. Bypass the node when you're just iterating on unrelated parts of the workflow.
• If you're on a transparent-background workflow for V5, note that NovelAI's V5 supports native transparent background generation - this node is for removing a background from an existing image, which is a different job.