RMBG 2.0 – Background Removal Model

What is RMBG 2.0?

Bria AI's latest flagship model, RMBG 2.0, is designed for professionally removing image background. It is trained on licensed data, making it safe for commercial use as it generates compliant outputs. The model is ideal for developers, creators, and product managers to get speedy background removal with precision and accuracy. It functions on deep learning to create smooth alpha channels with natural and clean transparency, with a soft transition. RMBG 2.0 utilizes non-binary masks with 256 transparency levels, ensuring harmonious and invisible image merging and preserving the fine details for realistic image output, unlike the commonly available models, which produce harsh edges with obvious binary masks. Furthermore, being a part of Bria AI’s image-editing suite, it operates with asynchronous processing for scalable image handling to manage and efficiently execute large volumes of background-removing tasks.

Key Features of RMBG 2.0

• High-precision background removal: It is designed to automatically detect foreground objects and create clean masks.
• preserve_alpha (default: true): This prompt can be used to retain transparency for semi-transparent edges; you also set "false" for solid backgrounds.
• **Non-binary masks **: Option to toggle precise edge thresholding for sharper, high-contrast masks.
• Asynchronous API: This makes it possible to execute high-volume workflows; poll for completion, and handle large-scale tasks without blocking.
• Content moderation toggles:
  • visual_input_content_moderation – to flag sensitive inputs before processing
  • visual_output_content_moderation – to ensure safe outputs for public use
• Smooth transparency: Anti-aliased edges create gradual, soft blending between the object and its background.
• Suite integration: It can be integrated with Bria AI generative fill to achieve background replacement or enhancement models.

Best Use Cases

• E-commerce & Retail: Possible to isolate products on white or transparent backgrounds to design catalogs and A/B tests.
• Graphic & UX Design: Useful to create clean assets for UI mockups, icons, thumbnails, and marketing collateral.
• Social Media Content: Save time by quickly swapping backgrounds or placeholders for banners, ads, and posts.
• Digital Art & Compositing: Easy to prepare foreground elements for collage, layering, or generative fill.
• Automated Pipelines: It can be integrated into CI/CD systems for automated image asset preparation.

Prompt Tips and Output Quality

1. Provide a valid image URI:
   “image”: “https://example.com/product.jpg” or Base64-encoded string.
2. Adjust preserve_alpha:
   • true – PNG with semi-transparent edges
   • false – JPEG-style opaquecutout
3. Enable moderation for public channels:
   • visual_input_content_moderation: true
   • visual_output_content_moderation: true
4. Use high-resolution inputs (≥1024 px) for cleaner masks.
5. Combine with custom binarization for logos or text:
   { "preserve_alpha": false, "custom_threshold": 0.5 }
6. For targeted edits, use Bria AI’s masking and generative fill models in sequence.

FAQs

Q: Which image formats are supported?
A: RMBG 2.0 supports JPEG, PNG, WebP via URI or Base64. Output supports PNG (with alpha) or JPEG (opaque).

Q: How do I retain transparency?
A: Set preserve_alpha: true; RMBG 2.0 will export a PNG with smooth, anti-aliased alpha channels.

Q: Can I process images in bulk?
A: Yes, you can use asynchronous endpoints to execute multiple jobs and poll for results programmatically.

Q: What is content moderation?
A: Two toggles ensure the output is "safe" for commercial use -
• visual_input_content_moderation: blocks sensitive inputs
• visual_output_content_moderation: filters outputs before public release.

Q: How accurate is the mask?
A: RMBG 2.0 achieves >95% pixel-level accuracy on high-contrast, well-lit inputs. The non-binary mask with 256 transparency levels further gives sharp edges.