RMBG v1.4
briaai/RMBG-1.4
published Dec 2023 · updated Jul 2025
RMBG v1.4 is a background removal segmentation model that separates foreground from background in images, trained on over 12,000 high-quality licensed images for commercial content creation.
specs
| Task | Background removal / image segmentation |
| Architecture | IS-Net (Dichotomous Image Segmentation) enhanced with proprietary training |
| Parameters | Not specified |
| License | Creative Commons (non-commercial); commercial license required from BRIA |
about this model
RMBG-1.4 is a background removal (saliency segmentation) model that separates foreground from background across general stock images, e-commerce, gaming, and advertising content. Developed by BRIA AI, it is trained exclusively on a professional-grade, legally licensed dataset of over 12,000 high-resolution, pixel-wise manually labeled images with balanced gender, ethnicity, and disability representation.
Key Strengths
- Built on the IS-Net architecture (ECCV 2022) with a proprietary training scheme and dataset, improving accuracy and effectiveness for diverse image types.
- Training data distribution: 87.7% photorealistic, 52.05% non-solid background, 51.42% single main foreground object; categories include objects only (45.11%), people with objects/animals (25.24%), people only (17.35%), and others.
- Designed for commercial use cases requiring content safety, legally licensed data, and bias mitigation.
Qualitative Results
Benchmark Context
The underlying IS-Net was evaluated on the DIS5K dataset (5,470 high-resolution images, up to 4K) against 16 segmentation models, introducing the Human Correction Efforts metric. RMBG-1.4 is trained on more than double the original dataset size (12,000+ images). A successor model, RMBG-2.0, reports further quality improvements.
Licensing & Compliance
Non-commercial use is permitted under a Creative Commons license (CC BY-NC 4.0 per the RMBG-2.0 card). Commercial use requires a separate license from BRIA. BRIA trains exclusively on licensed data from Getty Images, Alamy, and Envato, and holds SOC 2 Type II, ISO 27001, GDPR, and EU AI Act certifications.
best for
- ·Removing backgrounds from e-commerce product photos
- ·Isolating foreground objects in advertising content
- ·Extracting subjects from gaming and stock images
FAQ
No, it is licensed under Creative Commons for non-commercial use only. Commercial use requires a separate license from BRIA.
The model accepts standard image formats (e.g., JPEG, PNG) loaded as PIL images or numpy arrays; the pipeline returns a mask or a composited image.
Use the gigarouter OpenAI-compatible endpoint with your API key, specifying the model as briaai/RMBG-1.4 and sending an image URL or base64-encoded image.
It is based on the IS-Net architecture from the DIS paper (ECCV 2022), enhanced with a proprietary training scheme and dataset.
It was trained on over 12,000 high-resolution, manually labeled images from licensed sources including stock, e-commerce, gaming, and advertising.
We're benchmarking and onboarding RMBG v1.4 as a hosted, OpenAI-compatible API. Sign in for free credit and be ready when it lands, or tell us you want it and we'll prioritize it.