Now in Beta: Object Detection upgrade in Object Masking.

The object detection in Object Masking tool is getting an upgrade. If you are using the latest Beta, you are already using it. We wanted to share what's changing.

When you move your cursor over the Program Monitor with the Object Mask tool active, colored overlays snap to objects in the frame before you click. That instant preview is meant to be fast, confident, and complete. It works great for common foreground subjects. But for many other things you might have found yourself reaching for the rectangle or lasso fallback more often than you'd like.

What's changing

We've upgraded the underlying model powering object detection. The new experience is meaningfully better across three dimensions:

More objects detected. The new model surfaces roughly twice as many selectable objects per scene. Objects that were previously invisible to the object detection pipeline- props, background elements, layered subjects with props, now light up.

More complete object boundaries. Mask boundaries are more accurate on complex shapes, irregular edges, and subjects that bleed into similar-colored backgrounds. This means less lasso correction after selection.

Faster to a usable mask. The combination of a more capable model and async propagation (which shipped in 26.3) means faster tracking and better masks.

What to watch for in the beta

We did extensive on failure-prone footage, but there's no substitute for real-world editorial workflows. A few things we're specifically watching:

• Very small objects: Objects occupying a small portion of the frame. The new model handles these better than before, but the rectangle/lasso fallback remains your best path for anything close to the screen edge or below ~5% of frame area.
• Busy scenes with many objects: More detectable objects is a feature, but in extremely dense frames (think crowd shots or cluttered environments) we want to make sure the object detection experience stays snappy and doesn't feel overwhelming.
• Hardware limitation: The upgraded object detection experience requires hardware that meets a minimum capability. Intel-based Macs and Windows machines with AMD graphics cards running older drivers will continue to see the previous object detection model and won't benefit from this upgrade. If you're on one of these configurations, Object Masking still works, you just won't get the expanded detection coverage described above. For AMD GPU users on Windows, updating to the latest graphics drivers is the recommended first step; once updated, Premiere will pick up the new experience automatically.

If you're in the beta, try it on footage that previously frustrated you. Specifically:

1. Did you find yourself hovering over objects that used to be invisible?
2. Were there cases where the new model detected too many objects in a way that felt noisy or confusing?
3. Any scenes where the boundary felt clearly worse than before?

Feel free to drop your findings in the thread below. Screenshots and clip descriptions are incredibly helpful. If something looks off, note your OS, GPU, and the rough nature of the footage (interview, run-and-gun, studio, etc.).