Remove BG ✒️🅝🅐🅘

Background removal is the most commoditized operation in the AI-image world - BiRefNet is bundled into ComfyUI core and a dozen free nodes do it locally. So why would you reach for RemoveBGNAID, which routes the job to NovelAI's servers and charges you Anlas for it? Honest answer: for a couple of narrow cases, and otherwise you probably shouldn't. If you want free and local, close this page and use a BiRefNet node. But if you're already in NovelAI-land - the image is a NovelAI render, you're paying for the subscription anyway, and you want to keep the whole pipeline inside one pack without downloading a segmentation model - this does the job with three inputs and no local dependencies.

The inputs are the director-tool standard:

• image - what to cut out.
• limit_opus_free (on) - caps size at 1024×1024 to stay inside Opus-tier free generation quota.
• ignore_errors (off) - return a blank and keep the queue moving instead of failing loudly.

Output is an IMAGE with the background removed - and because the pack preserves the alpha channel when it decodes, you get a genuinely transparent PNG, not a flat checkerboard trick. That's the one thing worth stealing from this node: if you're building a pipeline that needs transparency and NovelAI's matting happens to be better on your specific anime render than the local default, the alpha comes out clean. The result is also autosaved to output/NAI_autosave/ with its original metadata.

Mechanically it's the same augment-image client as the other director tools: base64-encode, POST with req_type: "bg-removal", unzip, decode. NovelAI's model does the segmentation remotely; your GPU sits idle; each call burns Anlas. The KB's background-removal essay notes that "best" depends on which failure you can tolerate - hair, transparency, thin structures - and NovelAI's matting is tuned for anime illustration specifically, which is where it can genuinely outdo a generic local u2net. On photos, don't bother.

Downstream, the transparent output feeds naturally into compositing: put a LoadImage scene behind it, or pipe it through GenerateNAID as an img2img source where you control how much the background re-blends. Install is the pack standard - Manager search "ComfyUI_NAIDGenerator", or clone into custom_nodes/ and restart - with NAI_ACCESS_TOKEN set in a .env in the ComfyUI root. And remember the recurring gotcha: if you get a blank back with ignore_errors on, check the console for the swallowed 401 before assuming the model failed.