Bria RMBG

RmbgNode (shown as Bria RMBG) removes the background from an image and hands you back a cutout with real transparency. If you've spent any time in ComfyUI you already know this operation - BiRefNet is free and bundled in core, rembg is free, every pack has a version. So why would you pay Bria for it? Because this node does the job on Bria's servers. No model download, no VRAM, no dependency wrangling, and it runs on a laptop with integrated graphics. The tradeoff is that every cutout is a billed API call.

How it works

You feed images (an IMAGE tensor, or a batch of them) plus your api_key. The node base64-encodes each frame and POSTs to Bria's /v2/image/edit/remove_background endpoint, then polls until the result is ready and downloads the cutout. Because it loops over the batch, you can rip through a folder of product photos in one run. Each image in the batch gets its own API call and its own bill.

The preserve_alpha toggle (default on) is the one that matters most: it keeps genuine transparency in the result instead of flattening to a solid background, which is what you want if you're compositing the cutout onto a new scene later.

Inputs and outputs

• images - your input batch.
• api_key - the BRIA token, default placeholder BRIA_API_TOKEN (won't work until you replace it).
• preserve_alpha (default true) - keep RGBA output.
• visual_input_content_moderation / visual_output_content_moderation (both default false) - flip these on if you're processing sketchy content and want Bria's safety layer to look at it.

One output: output_images, a list of RGBA tensors. Note the failure behavior - if a single image in the batch errors, the node logs it and substitutes an empty transparent tensor of the same size rather than failing the whole run. That's a silent gotcha: check your results, because a blank cutout means that frame failed, not that the model found nothing.

Installing it

Part of the BRIA AI API nodes pack (Bria-AI/ComfyUI-BRIA-API). Install via ComfyUI Manager (search BRIA API) or:

cd ComfyUI/custom_nodes
git clone https://github.com/Bria-AI/ComfyUI-BRIA-API

Restart ComfyUI; the node lives under API Nodes. No heavy dependencies beyond requests (already present), no model files. You need a BRIA API token from platform.bria.ai and credit on the account.

When to use it vs. the free local stuff

Be honest about your use case. If you just need a clean cutout on your own machine, BiRefNet in core ComfyUI is free and, frankly, excellent on hair and semi-transparent edges - background removal is the most commoditized operation in this ecosystem. Where Bria wins is workflow: no model to load, consistent quality across resolutions, and the licensed-data angle matters if you're producing commercial assets and your legal team cares about provenance. Also the RMBG here is a solid matte - reliable edges, not just a "good enough" u2net-era cutout. Just remember the empty-tensor failure mode, and that the API polls on a timeout, so an unlucky batch takes as long as the slowest image.