How to Remove Image Backgrounds for Free with AI

The Old Way Was Expensive and Slow

For years, cutting a subject out of its background meant one of three things: paying for Photoshop and learning manual masking, uploading your photos to a paid service like remove.bg that charges per image after a free tier, or settling for crude color-key tools that fail the moment your subject shares a color with the background.

Try it free: Background Remover — Remove backgrounds from photos using AI. Runs on our server; the image is deleted after processing. No signup to try.

Each option had a cost — money, time, or quality. And all the cloud-based services had the same privacy problem: your image gets uploaded to someone else's server, processed there, and stored for some unspecified period. Fine for a stock photo. Less fine for a product prototype, a legal document, or a photo of your kid.

How AI Background Removal Actually Works

Modern background removers don't use color matching. They use a class of deep learning models trained on salient object detection — the task of identifying the most visually prominent object in an image and separating it from everything else.

Scanly's Background Remover uses an AI model based on the ISNet architecture (Dichotomous Image Segmentation), designed for high-accuracy cutouts. The model processes your image at 1024×1024 resolution through a multi-stage encoder-decoder network, producing a probability mask — a grayscale map where white pixels represent foreground (the subject) and black pixels represent background. Each pixel gets a probability score between 0 and 1, and the threshold slider determines the cutoff point. Compared to older segmentation models, ISNet is especially strong on hair, fur, semi-transparent fabrics, and fine edges.

The model was trained on over 10,000 images across diverse categories: people, animals, products, vehicles, text, and abstract objects. It learned to recognize object boundaries, handle semi-transparent regions like hair and glass, and distinguish foreground from background even when colors overlap.

💡 Key difference

Color-keying (green screen) removes pixels that match a specific color. AI segmentation removes pixels that belong to the background regardless of color. That's why the model can handle a white cat on a white couch — it understands object shape, not just hue.

Removing a Background — Step by Step

Open the Background Remover and upload an image. The tool accepts JPG, PNG, WebP, GIF, BMP, and AVIF files up to 50 MB. The image is sent to our server over an encrypted connection, where the AI model runs the cutout, and the file is deleted right after processing.

Step 1 — Upload. Drop your image or click to browse. It uploads over an encrypted connection and the model begins inference on our server. Processing takes a few seconds for most images. A progress indicator shows the current stage.

Step 2 — Review the result. The tool shows three views: the cutout result (subject on transparent or colored background), the raw segmentation mask (white = kept, black = removed), and a side-by-side comparison with the original.

Step 3 — Adjust if needed. Two sliders fine-tune the output. Threshold controls how aggressively the model separates foreground from background. Edge smoothing blurs the mask boundary for softer transitions. Click Re-run to apply changes.

Step 4 — Choose a background. Transparent (the default), white, black, or any custom color via the color picker. This is applied before download — no need for a separate editing step.

Step 5 — Download. The output is always PNG format (supports transparency). Resolution matches your input up to 2048 pixels on the longest side.

Want to try it? Remove the background from any photo in seconds — free, deleted after processing.

Remove Background Now →

When the AI Nails It — and When It Doesn't

The model works best when there's a clear visual distinction between subject and background. Product photos on solid backgrounds, portraits against blurred bokeh, animals on grass, objects on a desk — these produce clean cutouts with minimal adjustment.

It struggles in predictable situations. A subject that blends into the background (white shirt on white wall) gives the model ambiguous probability scores around the edges. Very cluttered scenes with no obvious single subject confuse the saliency detection — the model doesn't know what you want to keep. And fine details at the sub-pixel level — individual hair strands against a busy background — will always be approximated rather than perfectly traced.

These aren't flaws unique to this tool. Every automated background remover, including the paid ones, hits the same limitations. The difference is how much post-processing control you get. Photoshop lets you paint the mask manually pixel by pixel. Remove.bg gives you a binary result. Scanly gives you threshold and smoothing sliders, which cover most practical cases without requiring manual editing.

Threshold and Smoothing — What They Actually Do

The threshold slider (range 10–90, default 50) controls the probability cutoff. Every pixel in the segmentation mask has a score from 0.0 (definitely background) to 1.0 (definitely foreground). Setting threshold to 50 means pixels scoring above 0.5 are kept. Lower the threshold to 35 and you keep more borderline pixels — useful for preserving hair wisps, transparent objects, and soft shadows. Raise it to 65 and you get a tighter, cleaner cutout with sharper edges, but you may lose semi-transparent areas.

The edge smoothing slider (range 0–10, default 3) applies a multi-pass blur to the mask boundary. At 0, the edge follows the raw model output — often slightly jagged because the model predicts probabilities on a pixel grid. At 3–4, edges look natural for web and social media. At 7–10, the smoothing becomes visible as a soft glow around the subject, which can help blend into a new background but loses fine detail.

🔍 Settings by use case

Product photos: threshold 50, smoothing 2–3. Clean edges, no fringe. Portraits with hair: threshold 35–40, smoothing 4–5. Keeps more detail at the hairline. Text and logos: threshold 55–60, smoothing 1. Sharp, crisp edges.

The Privacy Argument

Every cloud-based background remover works the same way: your image goes to a server, gets processed by a GPU-accelerated model, and comes back. The server sees your image. Depending on the service, it may be logged, cached, used for model training, or stored for a retention period you never agreed to read about.

Scanly handles this differently from the services that monetize what you send them. Your image is uploaded over an encrypted connection, processed by the AI model on our server, and deleted right after the cutout is returned — within one hour at most. It is never stored long-term, logged, or used to train models, and there are no tracking cookies or advertising profiles attached to it. For most work that is the privacy that matters. For genuinely confidential material — contracts with visible signatures, unreleased product photos — the safest option is always a tool that runs fully offline, and any browser tool that sends an image anywhere, ours included, is not that.

If you want to verify that a processed image contains no residual metadata from the original, run it through the EXIF Checker. Canvas-rendered PNGs are typically metadata-free, but it doesn't hurt to confirm — especially before sharing publicly. For bulk metadata stripping, the EXIF Remover handles batches of up to 10 files.

Background Removal for Specific Tasks

E-commerce and product listings. Most marketplaces (Amazon, eBay, Etsy) require or strongly prefer product photos on white backgrounds. Remove the background, select white from the color picker, and download — no Photoshop layer flattening needed. For marketplace-ready sizing, run the result through the Image Resizer or the Social Media Resizer for platform-specific dimensions.

Social media content. Transparent PNGs are the building blocks of layered social graphics. Remove the background, then composite in Canva, Figma, or any design tool. If you need to compress the result for faster loading, note that PNG transparency survives compression but switching to JPEG does not — JPEG doesn't support alpha channels.

Presentations and documents. A subject on a transparent background drops cleanly into slides without the white rectangle that ruins every PowerPoint deck. Export as PNG, drag into your slide. If you're building a PDF, the Image to PDF tool handles the conversion.

ID and passport photos. Some visa applications require photos on specific background colors. Remove the original background, pick the required color (typically white or light gray), and download. Pair with the Image Cropper to match exact dimension requirements.

Common Questions

Can I remove backgrounds from multiple images at once? The Background Remover processes one image at a time. For batch workflows, remove each background individually and download the transparent PNGs. Dedicated batch background removal tools exist if you have many images to process in one go.

Why does my result have rough edges around hair or fur? Hair and fur are the hardest edges for any background removal model. Individual strands blend with the background at sub-pixel level. Try lowering the threshold to 35–40 and increasing edge smoothing to 5–6. The result won't match manual masking in Photoshop, but it's typically good enough for social media and web use.

What happens to my image after background removal? Your image is uploaded to our server over an encrypted connection, processed by the AI model, and deleted right after the cutout is returned — within one hour at most. It is never stored long-term, logged, or used to train models. The transparent PNG is generated from the result and sent back to your browser to download.

A Free Tool That Replaced a $240/Year Subscription

Photoshop's background removal is better. That's worth stating plainly — Adobe has more compute, more training data, and a manual refinement workflow that no automatic tool can match. But Photoshop costs $22.99/month. Remove.bg charges $0.20–$0.90 per image after the first free download. Scanly's Background Remover is free to try — no subscription, no per-image fee, no account. For the vast majority of background removal tasks — product photos, social content, quick cutouts — it produces results that are good enough, instantly, for free. Sometimes good enough at zero cost beats perfect at a subscription price.