BRIA Eraser

Positioning Overview

BRIA Eraser is an image editing model focused on erasing: it takes an original image and a black-and-white mask, where white marks the areas to be removed. The model removes the target objects and reconstructs the occluded pixels, outputting a clean image at the original resolution. It is suited for developers who need to embed object removal and temporary-element cleanup into batch content pipelines, especially e-commerce, advertising, and concept design teams that are sensitive to asset licensing provenance and require results that can be published directly or fed into post-production workflows.

Key Highlights

• Focused on erasing and region reconstruction — it does one thing: remove targets according to the mask and fill in the background. The workflow is simple and easy to integrate into existing image processing pipelines.
• Output preserves the original resolution — results keep the input image dimensions, so no resizing back is needed and outputs can go straight into downstream layout or publishing workflows.
• Usability-first input constraints — supports publicly accessible JPEG, PNG, and WEBP images plus black-and-white masks of the same aspect ratio, and can preserve the alpha channel via a parameter.

Capabilities

Mask-driven object erasing: use a black-and-white mask with the same aspect ratio as the original image to locate the objects to remove — white is the erase area and black is the keep area — making it suitable for precisely removing specified elements from a scene.

Pixel reconstruction of erased regions: performs content completion on erased areas and reconstructs background structure, so that removed objects do not leave obvious holes and manual retouching is reduced.

Output resolution alignment: output follows the input image resolution, and returned results can directly replace the original asset, eliminating a second scaling and alignment step.

Alpha channel preservation: when preserve_alpha is enabled, inputs with an alpha channel remain transparent in the result, which is suitable for cleaning up layered assets.

Mask source adaptation: mask_type distinguishes hand-drawn masks from masks produced by segmentation tools, making it easy to feed automatic segmentation results directly into the erase workflow.

Use Cases

• E-commerce operations removing clutter from product images — input a product hero image and a mask marking clutter and temporary elements to get a clean product shot that can be used directly on detail pages and store listings.
• Post-production cleanup of advertising assets — locally erase unwanted objects in campaign assets to get finished images ready for layout and delivery, reducing back-and-forth manual retouching.
• Local touch-ups in concept design — input a character concept art or scene sketch and a local mask to remove extra strokes or elements, producing a base image that can be further developed.
• Batch content processing pipelines — feed masks produced by automatic segmentation into the erase step in batches to get cleaned assets with uniform specifications, integrating with existing review and iteration workflows.

Limitations and Selection Advice

Results depend heavily on mask quality: the mask aspect ratio must match the original image, and edges should extend slightly beyond the target object, otherwise residual artifacts or accidental deletions are likely. The model only covers erasing and region completion and does not provide general-purpose image generation; scenarios that require generating images from scratch should use a different model. In addition, it only accepts publicly accessible image URLs, and results keep the original resolution; it is not suitable for strongly real-time interactive call scenarios.