BiRefNet v1 Base

BiRefNet quietly became this ecosystem's default background remover, and not because it set out to be one. The paper was about dichotomous image segmentation, but the same weights happen to be visibly better at the exact cases where rembg's old u2net had always been wrong - hair, fur, fabric mesh. By 2025, rembg itself bundles seven BiRefNet variants, and "just run BiRefNet" is the standard answer to "how do I cut this out cleanly."

Runware_birefnet_v1_base is the general-purpose flagship of that family, served over Runware's cloud. Feed it an image, get back the subject cut out. No model download, no VRAM, no pip install rembg adventure - the runware:112@1 checkpoint runs remotely and the node returns a native IMAGE.

How it works

The node sends a removeBackground task to the Runware API with your image (as a data URI). The model produces a segmentation and returns the foreground cutout, which the node downloads into an IMAGE tensor. Note the pack treats background removal as an image-producing task - the output is the actual cutout, ready to composite.

Inputs that matter

There is genuinely one input that matters:

• image - the IMAGE tensor you want cut out. Everything else is plumbing.
• outputFormat - default JPG. This is the trap: JPG has no alpha channel. If you're compositing the cutout onto anything, set PNG (or WEBP) or the transparency is flattened into whatever the background color is.
• outputQuality - 20–99, default 95; compression level for the saved output.

Install

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

Or ComfyUI Manager → search Runware → Install → restart. API key in ComfyUI Settings → Runware API key or RUNWARE_API_KEY. Prepaid top-up; the title bar shows the per-run cost, and background removal is cheap.

Gotchas

• Set PNG before you queue if you need the transparency. This is the #1 way people burn a run.
• Hair and fur are where BiRefNet shines, but flyaway strands are still the classic failure case - the KB's background-removal doc is blunt that no remover has won this outright (InSPyReNet still wins specific comparisons).
• The image goes up and comes back over the network, so very large inputs cost more and take longer. For bulk batch work, the pack's 512×512 fixed-resolution variant is the cheaper sibling.

If you only add one BiRefNet node, make it this one. It's the general-purpose cutout model, cloud-hosted, and the closest thing this pack has to a "just works" background remover.