BiRefNet background removal

About

BiRefNet removes backgrounds. There is no prompt and nothing to click: it finds the salient foreground on its own and returns either the matte for that subject or the subject already cut out on a transparent background.

Output dimensions always match the input. BiRefNet runs at a fixed 1024 internally and resizes the matte back to your source’s own width and height, which is also why the price is flat — the compute is the same whether you hand it a thumbnail or a full-resolution photograph.

The matte is soft rather than binary: edge pixels carry partial coverage, which is what keeps hair, fur and thin strokes intact instead of turning them into a jagged stencil. In one of our own tests it held the whiskers on a line drawing — the kind of detail that separates a usable cut-out from one you have to repaint by hand.

Set applyMask to get the RGBA cut-out instead of the bare matte. The alpha channel is the same matte either way, so the second form is strictly more useful when you want pixels rather than a mask.

BiRefNet is prompt-free by design, and that is the trade. When you need a specific object out of a busy frame rather than "whatever is in front", that is a different job and a different model.

One measured run on an RTX 5090 took about 1.0 second warm — a single run rather than a benchmark. In the Sogni app you reach it from any image you have generated or uploaded: open its menu and choose Remove background. It runs on the Sogni Supernet, a decentralized network of creator GPUs, so there is nothing to install.

Prompting tips

There is no prompt, so everything depends on the picture you hand it:

• One obvious subject — BiRefNet decides what the foreground is. A frame with a single clear subject gives it an easy decision; a crowded scene makes it guess.
• Leave a little room — A subject cropped hard against the frame edge has no background to separate from on that side. A small margin gives the matte somewhere to resolve.
• Separation beats resolution — Tonal or colour contrast between subject and background matters more than megapixels — the compute is identical at every input size anyway.
• Trust it with fine detail — Hair, fur, whiskers and thin strokes are what the soft matte exists for. Do not pre-crop them away in the hope of helping.
• Ask for the cut-out when you want pixels — Set applyMask to get an RGBA image instead of the bare matte. The alpha channel carries the same matte, so nothing is lost by taking the more useful form.

Matte or cut-out

One switch decides what comes back. Both forms are the same matte; the second one has already applied it.

Pricing

Use pay-as-you-go Spark packs for each render (1 Spark = $0.005), or choose a flat-rate Sogni plan for credit-free fair-use generation in the app.

One flat rate at any source size. BiRefNet always preprocesses to a fixed 1024 and resizes the matte back, so the compute per request is constant however large the image is.

1 Spark = $0.005. Pay as you go with Spark packs, or remove backgrounds under fair use on a flat monthly Sogni plan.

API

One Sogni API key reaches every model on the Supernet — call BiRefNet with the exact model id.

import { readFileSync } from 'node:fs';
import { SogniClient } from '@sogni-ai/sogni-client';

const client = await SogniClient.createInstance({
  appId: crypto.randomUUID(),
  apiKey: process.env.SOGNI_API_KEY,
  network: 'fast',
});

const project = await client.projects.create({
  type: 'image',
  modelId: 'birefnet_image_background_removal_fp16',
  positivePrompt: '',
  numberOfMedia: 1,
  startingImage: readFileSync('portrait.jpg'),
  applyMask: true, // RGBA cut-out instead of the bare matte
});

const [url] = await project.waitForCompletion();
console.log(url); // PNG at the source image's own size

import asyncio, os
from sogni_client import SogniClient

async def main():
    async with await SogniClient.create(
        api_key=os.environ["SOGNI_API_KEY"],
        app_id="birefnet-example",
    ) as sogni:
        project = await sogni.projects.create(
            type="image",
            model_id="birefnet_image_background_removal_fp16",
            positive_prompt="",
            number_of_media=1,
            starting_image="portrait.jpg",
            apply_mask=True,  # RGBA cut-out instead of the bare matte
        )
        for url in await project.wait_for_completion():
            print(url)  # PNG at the source image's own size

asyncio.run(main())

Call birefnet_image_background_removal_fp16 with a source image and no prompt. Background removal is not a Creative Agent tool, so there is no REST workflow alias — use the JavaScript or Python SDK. Full reference at docs.sogni.ai.