Create a clean transparent cutout from portraits, products and graphics

The Remove Background tool creates a cutout by making background pixels transparent while keeping the main subject visible. A transparent PNG is useful when a product photo needs to sit on a different page colour, when a portrait will be placed in a design, or when an object needs to be composed over another image. The tool accepts common JPG, JPEG, PNG and WebP files through its upload control. You can add more than one image and process a group rather than repeating the same upload steps for every photo.

The default Auto Detect mode first tries a neural background-removal model through the `@imgly/background-removal` package. This is the AI-based path intended to identify the foreground subject and separate it from the background. The model runs in the browser after its code and required model assets are available. If the neural path fails to initialize or process a file, the app falls back to a simpler colour-distance routine. That fallback estimates a background colour from samples around the image border and makes pixels with a similar RGB colour more transparent. It is not the same as a full subject-segmentation model and can remove matching colours that also appear inside the subject.

Two other detection modes are available: Solid Black and Solid White. These modes use the colour-distance process with black or white as the assumed background colour. They can be useful for an image photographed against a clean, uniform background. They are not a general-purpose replacement for selecting a complex background or separating hair from a detailed scene. The tolerance slider changes how much colour difference is treated as background. The visible control ranges from 10% to 70%, with 30% as the initial value. A higher threshold removes a wider range of similar colours but can also eat into subject edges.

The workspace includes a before-and-after comparison slider so you can inspect the original image and the processed result. The Preview Background selector lets you view the cutout against a transparent checkerboard-style preview, white or black. This setting changes the viewing background in the editor; the downloaded file remains a PNG with transparency. The preview colour is not painted into the PNG output.

After processing, a single image can be downloaded as a PNG with a `_nobg` filename suffix. When multiple images are selected, the processed PNG files are packaged in a ZIP archive. Each output keeps the dimensions returned by the image processing step. The tool does not provide a custom crop, resize or subject-position editor in this workflow. Inspect the edges and subject before downloading, particularly where the original background and foreground share similar colours.

A clear subject with a visible boundary is easier to separate than a subject that blends into the surroundings. Product photos on a plain background, portraits against a simple wall and objects with strong colour contrast are useful starting points. Hair, fur, glass, transparent packaging, thin wires, motion blur and shadows are more difficult because the subject edge contains fine details or colours close to the background. Always inspect small edge areas at a larger zoom before using the file in a product listing or design.

The Auto Detect model is the best starting choice for ordinary foreground subjects, but it can still leave unwanted pixels or remove details. If the neural model is unavailable, the fallback uses a sampled background colour and compares every pixel by RGB distance. It does not trace only a connected region from the image borders. As a result, an area inside the subject that happens to be close to the sampled background colour can also become transparent. The white and black modes use a fixed assumed background colour and depend heavily on a reasonably uniform source.

Tolerance controls how broad the colour match should be in the non-neural detection path. A lower setting is more conservative and can leave a fringe of background around the subject. A higher setting removes a wider colour range, which may clean up a halo but can also remove light clothing, reflective objects or highlights. Make small changes and compare the result rather than increasing the threshold to its maximum immediately. The visible slider uses 10% to 70%; the displayed percentage is a sensitivity setting, not a percentage of the image to delete.

The tool includes a feathered transition near the detection threshold. This softens some edges, but it is not a detailed manual hair brush, mask painter or restore/erase editor. If a cutout needs careful repair, use a dedicated masking editor with manual brush controls. This page also does not generate a new background scene or replace the transparent area with an arbitrary photograph. Its output is a transparent PNG.

The source file is processed in the browser. Auto mode imports a client-side background-removal package and may need to load model assets before the first successful cutout. The image itself is passed into that local processing function; the current tool does not send the source photo to a TinyImageResizer image-processing endpoint. Performance depends on the device's memory, browser capabilities, image dimensions and model startup. Very large images or long batches can take more time and may use significant device resources.

Before using the PNG:

This tool is a quick browser-based starting point for transparent cutouts. The result should always be reviewed, because AI segmentation and colour-distance removal can make mistakes on difficult images.