Bria

The new standard for video background removal

Live virtual backgrounds, real-time green-screen, AI-generated backplates, without the ML lift.Now powered by V-RMBG 3.0, bria’s new state-of-the-art model
Try it in the sandboxRead the API docs

One suite. Every endpoint you need.

What you get out of the box

Quality

Quality

Powered by V-RMBG 3.0, the latest in the RMBG family (5M+ Hugging Face downloads and 90% usable-result benchmark).

256-level alpha, high-res, clean on hair, motion, and semi-transparent objects.

Mask as output for full flexibility.

Latency

Latency

40 ms server inference at 24 FPS sustained.

<100 ms RTT in the US (p50 74 / p95 97).

Latest-wins frame scheduling drops stale frames.

Optimized design for live pipelines.

Scale

Scale

Hosted across cloud, BYOC, on-prem.

No GPU autoscaling, session management, or ML ops.

$0.0033/batch or streaming sec, pay-as-you-go

What teams are building with it

Blog posts

Branded retail environments, generated live
Live Commerce

Branded retail environments, generated live

Hosts stream from anywhere. The backdrop matches the campaign — studio one minute, lifestyle the next, flash sale the next. No physical set, no per-host setup cost, no green screen. Branded context follows the product, not the room.
A consistent identity across every stream
Creator · IRL Streaming

A consistent identity across every stream

Podcasters, vloggers, and IRL creators get a
controllable scene without leaving their actual
environment. Stylize, replace, or augment in real
time — without on-device ML, without a self-
hosted model, and without compromising the matte on hair or motion.
From overlays to immersive scenes
Game Streaming

From overlays to immersive scenes

Streamer backgrounds today are a static layer —
alerts, banners, branding piled on top of a
webcam feed. Bria replaces the layer with a
controllable scene. Match the game, swap on a
sponsor moment, react to chat. The backdrop
becomes part of the show.
Stage-quality broadcasts from anywhere
Live Events · Webinars

Stage-quality broadcasts from anywhere

Remote speakers join a hybrid event from home
offices and end up on the same virtual stage.
Backdrops switch per segment — keynote,
panel, sponsor break — without producer
scrambling or static green-screen rigs.
One platform, every studio
Education · Fitness · Coaching

One platform, every studio

Yoga from a living room becomes a sunset
beach. A coaching session from a kitchen
becomes a branded studio. Same hosts, same
equipment, consistent production value across
every session.
Avatars on customer-controlled scenes
Video Calls · AI Avatars

Avatars on customer-controlled scenes

Most avatar products composite onto static
backgrounds. Bria's streaming endpoint puts
avatars into reactive, customer-branded scenes
— at the matte quality and frame latency live
video actually demands. Same API for sales
calls, support sessions, and AI-driven product
demos.

Explore it before you write a single line of integration code

Free sandbox. Drop in a video URL or open a streaming session — the whole suite, your footage, no signup gate.

Try it in the sandbox

FAQs

Anything that needs the foreground separated from the background of a video. Live AI avatars on branded scenes, real-time virtual backgrounds for video conferencing, live commerce streams with reactive backdrops, live recording studios with clean-cutout preview, AR overlays. On the batch side: e-commerce product video, ad creative variations against dynamic backplates, talking-head localization across markets and backdrops, automated video editors. Same API key for all of it.

Bria builds visual AI infrastructure for product teams – image generation, image editing, and video editing – trained exclusively on licensed data. The video background removal suite is one capability inside a broader visual AI platform. Existing customers using Bria's image APIs can extend to video without changing infrastructure or auth.

40 ms server inference end-to-end. For US clients hitting US hosting, total RTT is <100 ms (p50 74 ms / p95 97 ms). Non-US clients hitting US hosting add network physics on top - use regional / private / BYOC / on-prem to close that gap. The endpoint sustains 24 FPS over a persistent WebSocket session, with latest-wins frame scheduling so stale frames are dropped rather than queued.

$0.14 per video second consumed, uniform across all four endpoints. Pay-as-you-go from the first call – no MSA, no minimum, no procurement step. Volume / regional / private / BYOC / on-prem pricing is a conversation; everything below that is on the meter.