About This Model
BiRefNet (Bilateral Reference Network) is a specialized image segmentation model designed for precise background removal. Unlike general-purpose image models that use prompt-based editing, BiRefNet is trained specifically for foreground-background separation, delivering cleaner edges and more accurate cutouts.
Key Features
- Precise edge detection — Handles hair, fur, semi-transparent objects, and complex boundaries
- Multiple model variants — General Use (fast/heavy), Portrait-optimized, and Matting modes
- Up to 2048×2048 — High-resolution processing for detailed cutouts
- Transparent PNG output — Clean alpha channel, ready for compositing
Best For
- E-commerce product photo background removal
- Portrait cutouts for ID photos, profiles, and compositing
- Design asset preparation (stickers, overlays, layered graphics)
- Batch background removal via API
FAQ
Which model variant should I choose?
General Use (Fast) works well for most images. Switch to Portrait for photos with people (better hair edge handling). Use General Use (Heavy) for complex scenes with fine details like fur or tree branches — it's slower but more accurate.
Why not use GPT Image 1.5 for background removal?
GPT Image 1.5 can edit backgrounds via prompts, but BiRefNet is a dedicated segmentation model — it produces cleaner cutouts with more precise edges, especially for hair and semi-transparent objects. It's also significantly faster and cheaper.