BackgroundErase is designed to preserve difficult details like hair, fur, thin edges, and transparency. These areas are the most challenging for any background removal model, and results can vary depending on the quality and composition of your input image.
Key idea: The better your input image, the better the model can preserve fine detail.
What counts as “fine detail”
Fine detail refers to areas where edges are soft, semi-transparent, or extremely thin. Common examples include:
- Hair strands and flyaways
- Animal fur
- Feathers
- Glass, smoke, or semi-transparent materials
- Thin wires or intricate objects
These regions require the model to estimate soft transitions rather than simple hard edges.
Best practices for better results
- Use a source-quality image (avoid screenshots, compressed images, or low resolution).
- Prefer lossless formats like PNG when possible.
- Keep the subject large and centered in the frame.
- Aim for an image that is close to a square aspect ratio—this improves model consistency.
- Ensure strong contrast between foreground and background.
- Avoid motion blur or heavy compression artifacts around edges.
Tip: Cropping your image closer to the subject (especially for portraits) can significantly improve hair detail.
Common issues and how to improve them
- Missing hair strands: Increase input resolution or use a higher-quality source image.
- Blurry edges: Avoid compressed formats (like low-quality JPGs).
- Edge halos or color bleed: Use API features like despill for cleanup.
- Chunky or blocky detail: Resize or crop the image to better frame the subject.