Blog

Guide

Bria Eraser: four mask-based editing examples

Bria Eraser costs $0.0575 per image. Explore four examples with source images and the matching input settings.

Oct 28, 2026 9

Per image: $0.0575

TL;DR

Bria Eraser: Upload an image and a black-and-white mask to erase selected areas. No prompt is needed.
Each image costs $0.0575, with one PNG per request. See pricing and budgets
The four examples include source images, masks and results. Open the model page and upload the matching inputs.

Prepare inputs for your edit

Eraser edits are controlled by the mask. White marks the area to erase and black marks what to keep. The scene descriptions explain the pictures; they are not generation prompts and must not be sent as a prompt parameter.

Street photo: remove a passer-by

Open the model page to upload inputs, then upload the matching mask; no prompt is required.

A passer-by in a long coat occupies the center of the source image, and the white mask covers the person. In the result, paving fills that area while the buildings remain visible on both sides.

Cover the entire person in white and check the shoes and outer edges. Keep the tables and chairs you want to retain in the black area.

This example is useful for inspecting how the background continues after the subject is removed. Paving fills the section previously hidden by the person. Follow the direction of the stones through the edited area and look for remnants near the former shoe positions. Visible buildings do not establish that every unselected pixel is unchanged.

For your own street image, start from the person's outline and check the complete transition to the road. If the person overlaps chairs, signs or other objects you need to keep, decide the acceptable editing boundary first. The original contains no view of the hidden paving, so the generated arrangement is not evidence of the exact original scene.

{
  "model": "bria-eraser",
  "input": {
    "image_url": "https://static.hiapi.ai/model-covers/2026/10/22/1792651217019-46c31ddb2b52110d-1792546025379-46c31ddb2b52110d-e1-street-src.png",
    "mask_url": "https://static.hiapi.ai/model-covers/2026/10/22/1792651218718-e5d2a17ae1d12d41-1792546025861-e5d2a17ae1d12d41-e1-street-mask.png",
    "mask_type": "manual",
    "preserve_alpha": false
  }
}

bria-eraser returns one image per request at $0.0575 per image.

Tabletop: remove paper and a tea bag

Open the model page to upload inputs, then upload the matching mask; no prompt is required.

The source has a paper ball and a tea bag to the right of a white mug. The mask covers both; the result keeps the mug and shows the wooden surface on its right.

Keep the mug in the black area and check that the white selection covers the edges of both unwanted objects.

The edit selects two separate targets while retaining the mug. Wood replaces the areas occupied by the clutter. Inspect the transition near the mug base and the grain direction, not just whether the paper is gone. Compare local brightness with the surrounding tabletop so a small preview does not hide an abrupt patch.

Cover each unwanted item separately. There is no need to paint over useful tabletop just to join the white regions. Check the selection boundary when clutter touches the mug. This example shows multiple targets in one mask; it does not establish equal results for any number of targets or every complex overlap.

{
  "model": "bria-eraser",
  "input": {
    "image_url": "https://static.hiapi.ai/model-covers/2026/10/22/1792651221006-aa722fbf0b20a05d-1792546026337-aa722fbf0b20a05d-e2-desk-src.png",
    "mask_url": "https://static.hiapi.ai/model-covers/2026/10/22/1792651221809-62ffea31d7a220c4-1792546026769-62ffea31d7a220c4-e2-desk-mask.png",
    "mask_type": "manual",
    "preserve_alpha": false
  }
}

bria-eraser returns one image per request at $0.0575 per image.

Interior: remove a cable and socket

Open the model page to upload inputs, then upload the matching mask; no prompt is required.

A vertical cable and socket appear on the wall above a sideboard. A narrow white mask follows them; the result shows a pale wall in that area, with the sideboard, plant and bowl still visible.

Trace the full cable in white and widen the selection around the socket. Keep the white mask away from the sideboard and plant.

The narrow cable and wider socket need differently sized selections. Their former location becomes pale wall in the result. Inspect the whole length for a band of different color or a repair boundary, then check the sideboard and plant against the intended edit.

Trace the cable end to end instead of selecting only its middle. Cover the socket's outline rather than extending the same narrow selection through it. This example has a relatively plain wall. Patterned wallpaper, reflections or intricate decorations require a fresh result check; this case establishes no common acceptance rate for them.

{
  "model": "bria-eraser",
  "input": {
    "image_url": "https://static.hiapi.ai/model-covers/2026/10/22/1792651224220-dc7497c3a8cb1fab-1792546027425-dc7497c3a8cb1fab-e3-room-src.png",
    "mask_url": "https://static.hiapi.ai/model-covers/2026/10/22/1792651225878-79d4af0aec50c743-1792546027894-79d4af0aec50c743-e3-room-mask.png",
    "mask_type": "manual",
    "preserve_alpha": false
  }
}

bria-eraser returns one image per request at $0.0575 per image.

Product photo: remove a sticker

Open the model page to upload inputs, then upload the matching mask; no prompt is required.

The sage-colored vase has a white sticker, covered by a rectangular mask. The sticker is no longer visible in the result. When preparing a transparent asset, check the alpha channel in the original PNG separately.

Set preserve_alpha=true when the input actually contains an alpha channel. Check the downloaded PNG for transparency; the background color on a web page does not establish it.

Where the sticker was removed, the result continues the vase's sage-colored surface. Look for a remaining rectangular boundary and check how color and shading meet the curved body. The silhouette and mouth are outside the removal target and deserve a separate comparison even when the sticker is gone.

The example retains preserve_alpha=true from the original request, but a web display does not establish that the output alpha channel fits your needs. Download the original PNG and inspect transparency and the outline before layout work. If the target is a product identifier that needs to remain, settle the delivery requirements first. This is a local-removal example, not proof that the real texture hidden under a sticker can be recovered.

{
  "model": "bria-eraser",
  "input": {
    "image_url": "https://static.hiapi.ai/model-covers/2026/10/22/1792651228144-b34d7f02cfca5c3e-1792546028494-b34d7f02cfca5c3e-e4-vase-transparent.png",
    "mask_url": "https://static.hiapi.ai/model-covers/2026/10/22/1792651229207-b89006e510782ca4-1792546028862-b89006e510782ca4-e4-vase-mask.png",
    "mask_type": "manual",
    "preserve_alpha": true
  }
}

bria-eraser returns one image per request at $0.0575 per image.

How should you adapt an example?

Identify the object to remove, then prepare a mask matching the source aspect ratio.
Extend the white area slightly beyond the object edge while avoiding subjects you want to keep.
Choose mask_type according to the mask source, and use preserve_alpha when the source is transparent.

Choose a similar problem before adapting the method

Match the target and background structure: the street for a person over paving, the tabletop for separate clutter, the cable for a narrow target, and the sticker for a local product edit. Similarity tells you what to inspect; it does not let you apply an example mask to a different image. A new source needs a matching selection.

For a reproduction attempt, use the paired source and mask and retain the complete request. Keeping settings but substituting your own images creates a different task. The browser flow needs no prompt, and the API request should not add one. The same inputs and settings express the same edit intent; this article promises no identical pixels on another run.

Inspect the source and mask before selecting a result by appearance. Record rejection reasons as specific regions: a remaining shoe edge, a discontinuity beside the mug or unsuitable transparency at the vase boundary. Then decide whether the next action is a mask correction or another candidate.

What can the displayed images establish?

The images let you inspect whether targets disappeared and how broad layout and texture continue. They do not establish pixel-identical preservation or recover hidden content as a verifiable record. Display resizing and image format also affect detail judgments, so inspect the original download for delivery.

The four examples cover only some materials, overlaps and boundaries, with no batch acceptance rate or completion-time measurement. Keep input preparation, task results and delivery acceptance separate. If alpha or product detail is a delivery requirement, check it directly; no sample image can substitute for that inspection.