Bria RMBG 3: background removal with clean edges on hair, fur and glass

Bria RMBG 3 removes image backgrounds with clean edges on hair, fur, glass and fine detail, returning a transparent PNG. Send one image, with no prompt.

Bria RMBG 3 removes an image's background and returns a transparent PNG — you send one image, with no prompt. Bria tuned it for the edges that defeat most removers: flyaway hair, animal fur, sheer fabric, glass and fine detail. It keeps partial transparency in the alpha channel, so a cutout drops onto a new background without a fringe. Against Bria's own RMBG 2.0, it trades sub-second speed for cleaner edges.

ModelBria RMBG 3Remove backgrounds with clean hair, fur and glass edges.

What it does well

Bria reports RMBG 3 ranked first among six background removers on 234 hard images. Each cutout below came from a single call, with no touch-ups.

Hair and fur

Loose hair and fur is where RMBG 3 stands out. Here a long-haired Maine Coon sits in front of a cluttered flower-market stall. The model keeps the whiskers, ear tufts and flyaway fur as a soft halo while the stall drops to transparency. About a quarter of the pixels end up as soft partial-alpha edges, so no feathering is needed before compositing.

OutputInputInputOutput
Drag the divider — whiskers and flyaway fur survive while the flower-market stall is gone.

Glass and other see-through objects

RMBG 3 does not flatten a transparent object into a solid silhouette. In this shot, a glass pitcher of lemon water and two clear glasses sit on a marble counter in front of a tiled wall. The counter and wall drop out while the pitcher, glasses and water stay see-through, so the checkerboard shows straight through the glass. That comes from the non-binary alpha, which stores hundreds of transparency levels instead of a hard on-or-off mask.

OutputInputInputOutput
The counter and tiled wall are gone, but the glass and water stay see-through.

Examples to try

One more hard input is sheer fabric. A peach organza scarf hangs on a wooden stand against a patterned wall.

OutputInputInputOutput
The wall drops out while the sheer organza keeps its weave and the wooden stand stays solid.

The wall is removed while the organza keeps its see-through folds and fine scalloped hem. The opaque wooden stand stays solid, so the non-binary alpha handles sheer and opaque regions in one pass.

Recommended settings

SettingValueWhen
preserve_alphatrue (default)Only affects inputs that already have transparency: on keeps it in the cutout, off discards it. No effect on an ordinary photo.
visual_input_content_moderationfalse (default)Set true to screen and reject flagged uploads in a user-facing pipeline.
visual_output_content_moderationfalse (default)Set true to screen the result before it is returned.

Good to know

  • RMBG 3 takes a few seconds per image. For sub-second removal, high-volume batches or prompt-guided cutouts, use RMBG 2.0 (bria-remove-background).
  • It accepts no prompt — sending one is rejected. To tell the model which object to keep with a prompt, use RMBG 2.0.
  • Quality starts with the input. A sharp, well-lit, high-resolution subject gives the cleanest edges; soft focus or heavy motion blur weakens the matte.
  • Send image as a public URL or a base64 data URI. JPEG, PNG and WEBP are accepted.

Use it from the API

RMBG 3 runs from one image. Send the input URL and keep preserve_alpha on:

APIRemove a backgroundAPI reference
# pip install "segmind>=1.1.0"
# export SEGMIND_API_KEY="YOUR_API_KEY"
import segmind

result = segmind.run(
    "bria-rmbg-3",
    image="https://segmind-resources.s3.amazonaws.com/input/bria-rmbg-3-input-66211e22.jpeg",
    preserve_alpha=True,
    visual_input_content_moderation=False,
    visual_output_content_moderation=False,
)
print(result.get("output"))
# 1. Submit the request; the reply carries its request_id
REQUEST_ID=$(curl -s -X POST "https://api.segmind.com/v2/bria-rmbg-3" \
  -H "x-api-key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "image": "https://segmind-resources.s3.amazonaws.com/input/bria-rmbg-3-input-66211e22.jpeg",
  "preserve_alpha": true,
  "visual_input_content_moderation": false,
  "visual_output_content_moderation": false
}' | jq -r .request_id)

# 2. Check its status until it is COMPLETED
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID/status" \
  -H "x-api-key: YOUR_API_KEY"

# 3. Fetch the result
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID" \
  -H "x-api-key: YOUR_API_KEY"

Resources

  • #image-to-image
  • #background-removal
  • #bria
  • #model-launch
  • #transparent-png
AM

Abhilash Mirji

@abhilashmirji

Product engineer at Segmind, shipping new AI models faster than my cats can knock things off the desk. 11+ years turning ambiguous problems into dependable products. Off-call: drumming, scuba diving, and getting outvoted by four cats and two dogs.

View profile

Posts

More from Abhilash Mirji

Make your own with Segmind

Run every model and Pixelflow in this post, and build your own workflows.

Start creating