Project 07, AI in the browser

Background Remover

Cut the subject out of any photo without uploading it anywhere. Drag the slider to compare, pick a new background, and save a PNG.

Runs on: Your CPU (WebAssembly)
Model: RMBG 1.4 by BRIA AI
First download: About 44 MB
Your browser: Checking

Remove a background

The model downloads on the first photo (about 44 MB) and is cached after that.

Drop a photo here
or click to choose. Paste works too.

Or try a sample:

New background

Waiting for a photo

Your result shows here. Drag the slider to compare with the original.

How it works

Find the subjectThe image is resized to 1024 by 1024 and the model returns a mask: white for the subject, black for the background.
Apply the maskThe mask is scaled back to full size and copied into the alpha channel of the original pixels.
ComposeThe cut-out is drawn over the background you picked, then exported as a PNG.

import { AutoModel, AutoProcessor, RawImage } from "https://cdn.jsdelivr.net/npm/@huggingface/transformers@3.8.1";

const model = await AutoModel.from_pretrained("briaai/RMBG-1.4", { config: { model_type: "custom" } });
const processor = await AutoProcessor.from_pretrained("briaai/RMBG-1.4", { config: {
  do_normalize: true, do_resize: true, do_rescale: true, rescale_factor: 1 / 255,
  image_mean: [0.5, 0.5, 0.5], image_std: [1, 1, 1], size: { width: 1024, height: 1024 },
  feature_extractor_type: "ImageFeatureExtractor", resample: 2, do_pad: false } });

const image = await RawImage.fromURL(url);
const { pixel_values } = await processor(image);
const { output } = await model({ input: pixel_values });

// Grayscale mask, same size as the photo: use it as the alpha channel
const mask = await RawImage.fromTensor(output[0].mul(255).to("uint8")).resize(image.width, image.height);