Remove Background Pixaroma

Cutting a subject out of an image is one of those tasks with a classic three-node chain: a Remove Background node, an Invert Mask node, and a Join Image with Alpha node to staple it back together. Remove Background Pixaroma collapses all of that into one node that hands you all three things at once - the transparent cutout, the foreground mask, and the inverted mask - plus it loads the model itself from a built-in dropdown. No separate model-loader node to wire, no mask math to get backwards. It runs on BiRefNet, which as of mid-2026 is the default answer for background removal in this ecosystem: MIT-licensed, sharp on hair and fur, and now even shipped in ComfyUI core.

How it works

Wire in your image, pick a model, run. The model processes at its internal resolution - 1024×1024 for the standard variant, 2048×2048 for anything whose filename contains hr or matt - then the mask is scaled back to your original size, so the output is the same dimensions as the input no matter which model you choose.

The dropdown reads ComfyUI/models/background_removal/ directly, and the filename controls which preprocessing resolution runs:

• birefnet.safetensors (Standard, ~424MB, 4–6GB VRAM) - the everyday cutout for clean objects, products, logos. Fast, and the right default.
• birefnet-hr.safetensors (HR, ~444MB, 8GB+ VRAM) - 2048px processing, for large images where fine outline detail matters (jewelry, intricate hardware).
• birefnet-matting.safetensors (Matting, ~444MB, 8GB+ VRAM) - soft alpha edges for hair, fur, lace, fabric. Matting predicts fractional transparency rather than a hard mask, which is why it's the one to try on flyaway strands and sheer material.

Downloads come from the Comfy-Org BiRefNet repo (standard) and ZhengPeng7's BiRefNet_HR / BiRefNet_HR-matting repos. The rename trap: HR and matting downloads come named model.safetensors. Rename them to birefnet-hr.safetensors / birefnet-matting.safetensors or they'll load - but run at 1024px, quietly losing the whole point. If the folder's empty, the node tells you exactly what to fetch and where to put it.

Outputs

• image - the cutout (RGBA) with the background transparent.
• mask - foreground in white, background in black.
• inverted_mask - the same mask flipped. This is the one you'd have built with an Invert Mask node yourself.

The honest take

For a hard-edged product cutout, Standard is fast and clean. For hair, you want the matting variant - and know going in that no background-removal model nails every semi-transparent edge, so budget for a little manual cleanup on the worst cases. It's also worth knowing the same BiRefNet files feed the AI background-removal buttons in the Composer and Paint editors, so one download powers three features. Install the pack, drop a model in models/background_removal/, and you're done:

cd ComfyUI/custom_nodes
git clone https://github.com/pixaroma/ComfyUI-Pixaroma

or ComfyUI Manager → Pixaroma → install, restart. No required Python dependencies; the BiRefNet path needs no extra pip packages. The usual update gotcha applies - node looks stale, hard-refresh the browser.