Bria Launches V-RMBG 3.0: State-of-the-Art Video Background Removal Model

New architecture delivers best-in-class temporal consistency, smarter foreground detection, and on-device support — with no integration changes required

TEL AVIV, Israel, June 10, 2026 /PRNewswire/ -- Bria, the Generative AI infrastructure for professional visual assets, today announced the launch of V-RMBG 3.0, its next-generation video background removal model. Built on a fundamentally new architecture, V-RMBG 3.0 processes video with an awareness of time, eliminating the flickering and unstable edges that have long been the visible ceiling of frame-by-frame approaches.

Background removal is an increasingly important capability in professional video pipelines, with the market projected to grow at a 15% CAGR through 2033. Avatar platforms, automated video editors, live recording studios, and live commerce platforms all depend on it to ship visual assets at scale. However, most models still struggle with fine details like hair, motion blur, fore/background distinction, and semi-transparent edges, which often leads to visible halos, flicker, and other artifacts in the final composite.

V-RMBG 3 solves those problems, treating each frame as a standalone image and using the previous frame's output as context when processing the next one. The model learns what the scene looks like over time, so it stops second-guessing itself from one frame to the next. The result is cleaner edges, stable masks across a full clip, and output that holds up on direct comparison against both open-source and commercial competitors.

Benchmarked head-to-head against every major competitor, V-RMBG 3.0 achieved a 63%+ win rate, preferred over every competitor tested on output quality. The speed advantage is just as decisive – up to 9x faster than every competitor tested. For teams running high-volume or real-time pipelines, the throughput difference is not marginal, it is the difference between a model that fits the workflow and one that bottlenecks it.

"The next wave of AI-native media isn't being built by humans reviewing individual frames, but by pipelines that run end-to-end without human correction loops," said Dr. Yair Adato, CEO of Bria. "V-RMBG 3.0 is our contribution to that shift: a model that understands time the way a human editor does, so the teams building tomorrow's content infrastructure don't have to work around it. Those are the standards we're aiming for across everything we ship."

The model has been benchmarked across +350 video clips covering the full range of production shot patterns — talking head, seated host, podcast setup, product demo, live commerce, and full-body — and validated against open-source models including MatAnyone 2, as well as commercial competitors Cutout and Veed.

Additional capabilities shipping with V-RMBG 3.0:

• Cloud, BYOC, on-prem, and on-device. The model runs wherever the customer needs it, including locally on their own machine. No cloud round trip, no data leaving the device, no per-second cost for local processing. Relevant for latency-bound real-time applications, offline and edge deployments, and privacy-sensitive content.
• Streaming-native. Processes frames one at a time, same as V2. The previous-frame context is the architectural improvement; autoregressive does not mean batching.
• Better video, no code change. Same API key, endpoints, parameters, and pricing. Customers on V2 get the upgrade automatically with zero integration work.
• Model weights and inference code available. Accessible to customers for self-hosted deployment, internal evaluation, and the training-data and runtime audits that closed-source APIs cannot support.

"Background removal opened the door to Bria for many of our production customers; V-RMBG 3.0 raises the ceiling on what that capability can do," said Yael Lubratzki Kurman, VP Product at Bria. "This is the right architectural fix for the problem, and because it ships as a drop-in upgrade, every customer on V2 gets better output the moment they make their next API call. No migration, no integration work, no trade-off."

V-RMBG 3.0 is available now through existing Bria batch and streaming endpoints, with no minimum spend or commitment required. Visit bria.ai to get started.

About Bria

Bria is the Generative AI infrastructure for professional visual assets. Controllable through Visual Generation Language. Rights-clear by design. Deployable wherever the work runs. Available as models, APIs, Skills/MCP, and end-to-end applications, built for builders, creatives, and agents shipping commercial work. For more information, visit bria.ai.

Media Contact

Sarah Small
Headline Media
sarah.small@headline.media
IL: +972 052 214 8601