Segmentation is the process of dividing an image or video frame into distinct regions or objects, identifying and labeling each pixel according to which subject, area, or category it belongs to. In AI imaging, segmentation creates a precise map of what is where in a frame, enabling tools to operate selectively on specific subjects, backgrounds, or regions without affecting the rest of the image.

There are several types of segmentation used in image and video AI workflows. Semantic segmentation labels each pixel with a category such as person, sky, road, or building, grouping all pixels of the same type together. Instance segmentation goes further, distinguishing between separate instances of the same category so that two people in a frame are identified as separate objects rather than a single "person" region. Panoptic segmentation combines both approaches, providing a complete scene understanding that labels every pixel with both its category and its individual instance identity. In practical AI tools, segmentation underlies background removal, selective masking, subject-aware inpainting, and the targeted application of effects or style changes to specific parts of an image. Video segmentation extends these capabilities across time, tracking segmented regions consistently from frame to frame as subjects and camera move.

Segmentation is a foundational capability that many visible AI generation and editing features depend on without surfacing the underlying technique directly to the user. When background removal identifies a person to separate from their environment, when inpainting fills only a selected region, or when a style effect is applied to one object while leaving others unchanged, segmentation is the technical process providing the per-pixel understanding that makes these selective operations possible.