NovelAI Declutter Preprocessor

The name tells you what it does: hand it a busy, noisy image and it strips out the clutter - stray marks, background noise, visual mess - while trying to leave your actual subject alone. This is a wrapper around one of NovelAI's "Director Tools," the suite of image-editing endpoints NAI built on top of their NAI Diffusion V3 base model rather than as a separate product. If you've got a sketch that's covered in construction lines, an image with distracting background detail, or a scan with dust and artifacts you don't want to hand-mask out, this is the tool that's meant to clean it up automatically before you do anything else with it.

How it works

Same pattern as every other node in this pack: it's not local processing, it's an API call. Your image goes to NovelAI's servers, their Declutter tool runs against it, and you get a cleaned-up image back. There's no local model, no VRAM cost - the only requirement is a working API key and an active NovelAI subscription behind it, since this endpoint is billed the same as any other NAI generation.

The inputs and outputs that matter

This node is about as simple as it gets - one input, one output, no knobs to turn:

• input_image (IMAGE, required) - the image you want decluttered.

Output is a single IMAGE - the cleaned result. From here it's a normal ComfyUI image: save it directly, feed it into the NovelAI node's own input_image for an img2img pass on the cleaned version, or hand it off to any other node downstream.

Because there are no exposed parameters, you don't get to control how much decluttering happens or dial in a threshold - you get NAI's tool as-is. If the result over-simplifies or under-simplifies your image, there's no local knob to fix that; the adjustment has to happen on NovelAI's side, if at all.

Installing it

Same pack as the rest of the NovelAI nodes here. ComfyUI Manager: search comfyui_nai_api, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/Creeper-MZ/comfyui_nai_api

Restart ComfyUI, then set your NovelAI API key in the pack's config.py. This node calls the API on every use, so without a valid key it'll fail the same way the generate node does - the README doesn't offer any more detail than "configure your API key in config.py," so that file is the one thing to get right before anything here works.

Common issues & troubleshooting

Auth errors are the first thing to check if this node fails outright - a missing, wrong, or expired key in config.py, same root cause as every other node in the pack.

Nothing visibly changes - if your input image was already fairly clean, don't expect a dramatic before/after. This tool is built for genuinely cluttered inputs; running it on something already tidy is a wasted API call more than a bug.

No control over strength - since the schema exposes only the input image and nothing else, you can't back off an overly aggressive result. If it's removing detail you wanted kept, your options are to pre-edit the image so less needs decluttering, or skip this node and handle cleanup manually with inpainting instead.

It's still a billed API call - same as generation, running this against a large batch of images adds up against your NovelAI account. Treat it as a deliberate step in your pipeline, not something to loop over freely.