How to Remove a Background from Any Image

How AI background removal works

Traditional background removal relied on colour thresholding or manual selection — techniques that struggled with hair, fur, or subjects that blended with the background. Modern tools use a neural network architecture called a semantic segmentation model.

The model is trained on millions of labelled images and learns to predict, for each pixel, whether it belongs to the foreground subject or the background. The prediction is a probability mask — a greyscale image where white means "keep" and black means "remove". The mask is then applied to your original image to produce a transparent PNG.

AIKitts uses the @imgly/background-removal library, which runs the ONNX model directly in your browser via WebAssembly. Your image never leaves your device.

What kinds of images work best?

AI segmentation performs best when there is clear visual contrast between the subject and its background. Results are excellent for:

• Product photos on plain studio backgrounds (white, grey, gradient)
• Portrait photos with a single person and a reasonably uniform background
• Pet photos where the animal is clearly the focal point
• Logos and icons on solid colour backgrounds

Results are more variable for:

• Images with complex, detailed backgrounds (foliage, crowds, busy urban scenes)
• Subjects with translucent edges (glass objects, flowing hair in wind)
• Low-contrast photos where subject and background share similar colours

Getting the best results

A few simple steps dramatically improve output quality:

• Use the highest resolution photo available. The model has more pixels to work with and the output mask is sharper. A 2000px wide image will give better edge detail than a 400px thumbnail.
• Avoid heavy JPEG compression on the input. Compression artefacts confuse the model at fine edges like hair. If you have an original PNG or RAW export, use that.
• Crop before removing. If your subject takes up only a small part of a large image, crop first so the subject fills more of the frame. This concentrates the model's attention on what matters.
• Check the mask before exporting. Zoom in on edge areas — hair, fur, fine fabric. If you see fringing or missed sections, a minor brightness/contrast tweak on the source image can sometimes help before re-running.

Common uses for a transparent PNG

• Placing a product image onto a website with a white or coloured background
• Creating profile photos with a custom branded background
• Building composite images in Canva, Photoshop, or Figma
• Preparing assets for print-on-demand stores (T-shirts, stickers, mugs)
• Removing distracting clutter from a photo before sharing on social media

How to use the AIKitts Background Remover

1. Open the Background Remover tool.
2. Click Select Image or drag your photo onto the tool. JPG, PNG, and WebP are all supported.
3. The AI model loads once (around 3–5 seconds on the first use) and then processes the image in your browser.
4. Preview the result and click Download PNG to save the transparent image.

The first processing run takes a few extra seconds because the ONNX model must load into memory. Subsequent images on the same page process almost instantly.