Bria Remove Video Background
The hosted cutter for moving subjects
- video
- VIDEO
Background removal is the most commoditized operation in this whole ecosystem for images - every model from rembg to BiRefNet is free and local. But do it to a video and you suddenly need per-frame segmentation that holds across 30 frames of movement, and that's a different problem, usually requiring a heavy local model or a lot of patience. This node skips both: Bria's hosted video background removal, exposed as a partner/video API node built into ComfyUI core. No install, nothing runs on your GPU; the cutout happens on Bria's servers through Comfy's proxy, billed per second against your Comfy account.
The interface is comically small for what it does. video in, background_color out - a combo offering ten solid choices from Black and White through the primaries and Orange. That's the whole creative surface, plus a seed that does the family-standard "re-run" duty on a non-deterministic pipeline. The node takes your clip, sends it to Bria's RMBG video endpoint, polls while it processes, and returns a VIDEO with the background replaced by the flat color you picked.
Here's the conceptual thing to grasp, and it's the source of most first-run confusion: the output is not transparent. You picked a background_color for a reason - the result is an opaque video on that solid color, delivered as MP4. If you actually want transparency, that's the sibling Bria Transparent Video Background node, which saves the alpha channel to a WEBM. Want green-screen instead? There's a Bria node for that too. This node is the "solid color" lane, and it's the right lane for compositing the subject onto a background you'll add later, or for a clean plate over a brand color.
Input limits worth respecting: clips up to 60 seconds are accepted, and the whole operation is async, so a longer clip means a longer wait while the node polls. On the quality front, expect the same edge-behavior tradeoffs the local image models trained you on - fine hair and semi-transparent materials are the eternal struggle, and video only makes them harder. Bria's hosted model is good, but if a clip comes back with a mangled edge, that's the category, not a misconfiguration.
The pricing shape is the other thing to internalize: it's billed per second of video, so a 30-second clip costs you many times a 5-second clip even though the per-call overhead is the same. Cut your source to what you actually need before you send it. And as with every hosted node, the video leaves your machine for Bria's infrastructure - keep that in mind for anything sensitive.
Bria's background nodes landed in core in early 2026, with the video variant following through the spring. If you need a fast, reliable subject cutout on a moving clip and don't want to babysit a local segmentation stack, it's the cleanest one-click version of the job in ComfyUI.
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| video | VIDEO | — | |
| background_color | COMBO | Background color for the output video. | |
| seed | INT | 00–2147483647 | Seed controls whether the node should re-run; results are non-deterministic regardless of seed. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| VIDEO | VIDEO | — |