Best Background Removal

BiRefNet Portrait is tuned for human subject segmentation, focusing on accurate extraction of faces and bodies with clean contours for portrait-centric workflows.

BiRefNet Matting extends binary segmentation into continuous alpha prediction, enabling smooth transitions for fine structures such as hair, fur, and semi-transparent regions.

Bria RMBG 2.0 is a refined background removal model designed for high-quality alpha matting and smooth edge transitions. It delivers accurate foreground isolation with subtle detail preservation, making it suitable for media production, compositing, and creative workflows that require clean cutouts.

BiRefNet HRSOD DHU is designed for salient object detection in high-resolution imagery. It preserves fine details and sharp boundaries, making it suitable for complex scenes with small or visually dominant subjects.

BiRefNet v1 Base is the core model of the BiRefNet family, designed for dichotomous image segmentation. It produces clean foreground–background masks with strong boundary precision by leveraging bilateral reference cues that preserve spatial structure and object integrity across diverse scenes.

BiRefNet General is trained for broad segmentation use cases, providing stable and consistent object masks across a wide variety of image domains without task-specific specialization.

This model is trained on multiple large-scale datasets, including DIS5K, to improve robustness across object scales, backgrounds, and visual distributions. It prioritizes generalization and consistency in real-world imagery.

This BiRefNet variant is specialized for camouflaged object detection, targeting objects that closely blend with their surroundings. It enhances subtle boundary contrast and improves detection reliability in low-visibility and texture-confusing scenarios.

BiRefNet Dis focuses on binary object segmentation tasks where clear separation between subject and background is required. The model emphasizes structural consistency and minimizes foreground leakage through dual reference guidance.

This variant is optimized for efficient inference at a fixed 512×512 resolution using FP16 precision. It balances segmentation quality with lower memory usage and faster runtime for production environments.

RemBG v1.4 is an optimized image background removal model focused on rapid and accurate separation of foreground subjects from background. It uses lightweight segmentation to produce clean alpha masks for visual assets, product photos, and portraits without manual editing.

Bria RMBG v2.0 removes complex backgrounds and produces clean alpha mattes for product photos and portraits. It supports varied subjects and lighting. Ideal for ecommerce pipelines and design tools that need fast accurate subject isolation at scale.