Bria RMBG v2.0

Here's the rare ComfyUI node with zero knobs that matter: one required input (image), one output (IMAGE), and the entire task is "remove the background." Bria RMBG v2.0 is the commercial-grade background remover that shows up in the removal-arena comparisons, known for handling hair and tricky edges well, and in this pack it runs entirely on Runware's servers. You don't download a checkpoint, you don't pick a variant, you don't fuss over input resolution. You wire an image in and a cutout comes back.

Background removal is famously the most-settled, least-settled operation in the ecosystem - everyone has a favorite local model (BiRefNet is the current default, InSPyReNet is the scrappy rival), and nobody agrees on a winner because there's no shared benchmark. So why would you use an API version? Three honest reasons: you're on a machine where even BiRefNet's modest VRAM is a problem; you're processing volume and want it off your GPU; or you're building a product and want one consistent, maintained endpoint instead of maintaining model files. If none of those apply, the local BiRefNet that ships in ComfyUI core is free and excellent, and you should use it. If you want this particular node, it's because the cloud trade is worth it for you.

Inputs and output

• image - required IMAGE. Anything with a subject.
• providerSettings.bria.preserveAlpha - default off. Only affects images that already carry an alpha channel; turn it on to keep it.
• outputFormat - JPG/PNG/WEBP, default JPG. This is the one setting you should actually change: a cutout in JPG gets flattened against a background the API picks and you lose your transparency. Set PNG if you intend to composite the result. That's the entire difference between "this node gives me a white box" and "this node gives me a usable cutout."
• outputQuality, ttl - leave them.

Output is a single IMAGE socket. Wire it into Save Image (as PNG), or straight into a composite/background-replacement workflow.

Install

One-time setup for the whole Runware pack:

cd ComfyUI/custom_nodes
git clone https://github.com/Runware/ComfyUI-Runware
pip install -r ComfyUI-Runware/requirements.txt

or ComfyUI Manager → search Runware, then restart and set your Runware API key (Settings → Runware API key, or RUNWARE_API_KEY). Dependencies are just runware-sdk, pillow, soundfile - no model files, because inference is remote.

The gotcha that actually bites

Beyond the JPG-flattens-alpha thing (which will get you once and then you'll set PNG forever), the real thing to internalize is that you're paying per cutout. For the occasional image that's fine; for a 500-image batch, a local BiRefNet pass is free and this is not. The node does report cost on its title bar after each run, so it's not a surprise - it's a prompt. Use this one for the images that matter, or for the volume you genuinely can't run locally.