Remove Image Background Online (AI BiRefNet)

Remove Image Backgrounds: AI and a JS Tool

You don't need to open a complex graphics editor and manually trace every outline to remove a background from a photo. FillTools offers two ways to process images: automatic background removal using the BiRefNet neural network and a classic JavaScript algorithm that runs directly in your browser.

These modes solve a similar task but use different approaches. AI analyzes the image and identifies the main subject, making it especially useful for photos of people, products, and complex scenes. JS works with pixel colors, so it's well suited for solid backgrounds, icons, diagrams, and screenshots.

After processing, you can review the mask, adjust the edges, erase unwanted areas, or restore parts of the image. The result is saved as a PNG. The choice of mode depends on the image itself: sometimes it's better to let the neural network handle the selection, and sometimes it's faster to use the regular color-based algorithm.

How AI and JS Modes Differ

Before you start, consider not only the quality of automatic selection but also the image type, privacy, and processing method. Below are the main differences between the two FillTools modes.

For a photo of a person against a room or street background, it's usually best to start with AI. If you need to make a uniform background transparent around a logo, diagram, or icon, try JS. This choice helps avoid unnecessary processing and yields a result that needs fewer manual corrections.

AI Background Removal with BiRefNet

The AI mode uses the BiRefNet-general (ONNX) neural network model. Instead of searching for a single background color, the model analyzes the entire image and generates a mask that determines which areas belong to the foreground and which should become transparent.

This matters for photos where there's no clear boundary between the subject and background. For example, hair may partially blend with surrounding details, fur creates an uneven outline, and clothing edges may be blurred due to lighting or motion. Simple color-based selection in such cases often leaves halos or removes part of the subject itself.

BiRefNet-general generates a foreground mask that can be used as a basis for the alpha channel and further adjustment of edge transitions. This is useful for portraits, pet photos, product shots, and other images where preserving a complex contour is important.

However, the neural network doesn't always correctly identify the purpose of each detail. A solid-colored sector of a chart, a thin line in a diagram, or part of a logo may be perceived as background. If your main goal is to accurately preserve the geometry and all graphic elements, the classic JS mode may be more practical.

For AI processing, the file is sent to a server where ONNX model inference is performed. After receiving the response, further mask adjustment happens on the client. According to the intended design, the service does not store the source file after the request completes. An internet connection is required, and repeated AI requests are limited by a set quota.

Classic Background Removal in the Browser with JS

Not every image needs a neural network. If the background has a uniform color, it can often be removed with a simple algorithm that analyzes pixel colors. In FillTools, this mode runs on JavaScript in the browser and uses Canvas to process the image.

An important advantage of this approach is that the photo is not sent to a server for background removal. Processing happens locally in the browser, without AI requests and the associated rate limits. This is a convenient option for simple images and for cases where you don't want to send the photo to an external server.

The JS mode offers two ways to determine the background color.

Selecting the background from the image edges

If you don't use the eyedropper, the algorithm starts checking from the image edges and looks for neighboring pixels whose colors are close enough to the colors along the perimeter. Areas of the background that are connected to the image edge and match the set sensitivity become transparent.

This principle is called flood fill. It's useful when you need to remove the background around a subject but not affect similarly colored details inside it, as long as they aren't connected to the outer background.

For example, this method can work well for a black icon on a white background or a dark object on a uniform light background. If the background has a gradient, shadows, or noticeable lighting variations, you'll need to adjust sensitivity, and the result may be uneven.

Selecting with the eyedropper

When you pick a color with the eyedropper, the algorithm switches to the chroma key principle. All pixels whose color is close enough to the selected one become transparent, regardless of where they are located in the image.

This differs from edge-based selection: the algorithm is no longer limited to the outer area connected to the perimeter. This approach can be convenient for graphics, screenshots, or images where the same background color appears in several separate areas.

At the same time, keep in mind that the eyedropper doesn't understand which parts of the image are the subject and which are the background. If the selected color is present in the subject itself, those pixels may also become transparent. Sensitivity determines how much colors can differ from the selected one to be included in the removal.

Mask Editor Features

Regardless of which mode you choose, after background removal you can switch to the shared editor and refine the result to your liking. You don't have to reprocess the entire image if the error affects only a small area.

• Brush. Use it for manual mask adjustment. You can change the brush size depending on whether you need to fix a small detail or process a larger area.
• Eraser. Remove unwanted visible fragments left after automatic selection or sensitivity adjustment.
• Restore. Bring back parts of the image that were mistakenly made transparent. This is especially useful for thin elements, object edges, and clothing details.
• Mask threshold. For AI results, adjust the boundary between visible and transparent areas. Changing the threshold can remove background remnants, but too aggressive a setting can also affect fine details of the subject.
• Edge softness. Smooth transitions between the subject and the transparent area if the contour looks too sharp. For graphics with crisp edges, a harder edge may be more appropriate.
• Invert mask. Swap visible and transparent areas if you want to keep the background and make the subject itself transparent.
• Background replacement. Keep the transparent background or choose white, black, or another color for the final image.
• Crop transparent margins. Remove empty space around the subject to make the PNG easier to use in layouts, catalogs, and other materials.
• Zoom. Enlarge the image for more precise work on contours, then zoom out to evaluate the result as a whole.
• History. Undo or redo actions during editing without starting the whole process over.

After automatic background removal, it's helpful to check the photo against a contrasting background. On a white background, light remnants may be almost invisible, while on a dark one, dark halos may show. If the contour needs refinement, use the manual tools before saving the final version.

What Tasks This Tool Is Suited For

Product photos for online stores

When filling a catalog, photos often come from different suppliers and vary in background, lighting, and quality. Background removal helps make product cards look more consistent, prepare images on a white background, or save the product as a transparent PNG.

For photos with shadows, highlights, and complex contours, it's best to start with AI. If you need to remove a uniform background from a simple product image, you can try JS. The resulting files can be used in online stores, catalogs, and advertising materials, but before publishing, check the specific platform's image requirements.

Logos, icons, and diagrams

For flat graphics, neural network selection isn't always the optimal solution. AI may perceive part of a solid-colored element as background, especially if you need to preserve all lines, sectors, or small details.

If the graphic is on a uniform background, JS mode often gives more predictable results. Edge-based selection helps remove the outer background, and the eyedropper makes pixels of the selected color transparent throughout the image. For more complex cases, you can adjust sensitivity or refine the mask manually.

Website and advertising design

Transparent images are convenient to use in banners, presentations, covers, service cards, and page layouts. A cut-out subject can be placed over another photo, a gradient, or a brand color without a rectangular background around it.

For photos of people and objects with complex contours, try AI. For interface screenshots, icons, and graphic elements with a solid background, JS is often more appropriate. After processing, you can save a transparent PNG and use it in Figma, Photoshop, or another editor.

Social media content

When creating covers, ad posts, stickers, and previews, you often need to separate a person or object from the original background. AI helps automate portrait selection, while JS can be useful for graphic elements with a clear solid background.

If the automatic result doesn't quite match your vision, you can edit the mask, change the background color, or crop transparent margins. This lets you prepare the image for composition without immediately resorting to complex manual selection.

How to Remove a Background from a Photo

1. Upload the image. Click "Choose file" and add the photo you need. For AI processing, the file will be sent to a server, while JS mode processes the image locally in the browser.
2. Select a mode. Enable AI for photos with complex backgrounds, people, products, and hair. Choose JS for solid backgrounds, icons, diagrams, or screenshots.
3. Click "Remove background". In AI mode, wait for the server response. In JS mode, adjust sensitivity and, if necessary, use the eyedropper to select the color to remove.
4. Check the mask. Make sure the parts of the subject you need remain visible and the unwanted background has become transparent. Zoom in if you need to inspect small details.
5. Fix any issues. Use the brush, eraser, or restore. For AI, you can additionally adjust the threshold and edge softness.
6. Choose a background and save the result. Keep transparency or set the desired color. If needed, crop empty margins and download the finished PNG.

How to Get the Best Result

The best processing method depends on what the image looks like. If the background is uniform and the subject's edges are clear, start with JS. If the photo contains hair, fur, complex lighting, or many small details, try AI. You don't need to use the neural network for every file; sometimes a simple color-based algorithm gives a cleaner result.

In JS mode, start with moderate sensitivity. If the background isn't fully removed, gradually increase it while checking that parts of the subject don't disappear. If a similar color appears in different places in the image, the eyedropper may remove it not only from the outer background but also inside the subject.

In AI mode, pay attention to fine details and semi-transparent edges. If part of the subject has become transparent, try changing the mask threshold or restoring it manually. If a halo remains around the object, check the edge softness and evaluate the result against a different background color.

For any mode, it's best to use an image of sufficient quality. Heavy compression, blurred edges, and similar colors of the subject and background can make selection harder. Before saving the final version, check the contour at full zoom and make sure the transparent areas truly match your expectations.

Choose the Right Method and Continue Your Work

Background removal doesn't always require the same approach. The BiRefNet neural network helps separate subjects in complex photos, while the classic JS algorithm lets you quickly remove a solid background without sending the image to a server.

In FillTools, both modes are combined with a shared mask editor. You can review the result, manually fix the contour, replace the background, or crop transparent margins, then save the image as a PNG. Choose the method that best suits your image and use the finished file for product catalogs, design, presentations, or social media posts.

AI mode requires an internet connection and has a request rate limit: one request per 60 seconds per user. JS processing doesn't use this AI limit because it runs in the browser.