Drop, paste, or browse one PNG, JPEG, WebP, GIF, SVG, AVIF, or other image this browser can decode.
{{ sourceName }}
{{ dropzoneTitle }}
Transparent PNG output is generated in this browser tab
Ignored {{ ignoredCount }} extra file(s). This tool processes one image at a time.
Choose the background evidence the browser should trust for the transparent PNG mask.
Use a color from the backdrop, not from the subject. Similar subject colors can become transparent.
{{ sensitivity }}%
Raise for uneven paper, walls, and shadows; lower when subject colors are close to the background.
{{ edgeSoftness }} px
Use 0 for hard icons, 2-4 for product photos, and higher only when the backdrop is very smooth.
Use a higher ceiling for final assets; lower it when a large photo makes the tab feel slow.
Switch the preview matte to catch white halos, dark halos, and semi-transparent edge residue.
Metric Value Detail Copy
{{ row.label }} {{ row.value }} {{ row.detail }}
Check Status Detail Copy
{{ row.label }} {{ row.value }} {{ row.detail }}
Tags: File Tools, Image
Image Metadata Stripper Strip image metadata in your browser, re-export a clean JPEG, PNG, or WebP, and compare privacy markers, GPS hints, size, and warnings.
Image Compressor Compress an image in your browser, resize or reformat it, compare size and quality changes, and strip supported EXIF before download.
Image Resizer Resize one image in your browser, compare fit, contain, cover, no-upscale, quality, and size-cap outcomes before downloading the file.
Image Format Converter Convert a single image locally into PNG, JPEG, WebP, AVIF, BMP, or ICO with resize controls, transparency handling, and byte-size warnings.
Customize
Advanced
:

A convincing cutout starts with the boundary between subject pixels and backdrop pixels. Background removal changes the alpha channel, the transparency value stored for each pixel, so a product, logo, or object can sit on a new page color, checkerboard, mockup, or catalog layout without carrying the original wall, paper, or screen behind it.

Simple background removal works when the background provides a trustworthy clue. A product on a paper sweep may have a pale border that can be sampled. A logo on a flat green or blue field can be separated by a color key. A catalog photo on a white studio surface may be bright and low in color variation. The risk in each case is the edge: if the subject shares the same color family as the backdrop, a color rule can remove the wrong pixels.

A simple image background is classified as transparent while subject pixels remain visible
Source Situation Why It Can Work Main Caution
Product photo on a clean sweep The border often represents the background color and shadow family. Gray products and soft shadows can be mistaken for backdrop.
Logo or object on a flat color A color key can mark matching pixels as transparent. Subject areas that share the key color may disappear.
White studio image Bright, low-chroma pixels usually separate from saturated subject color. Off-white fabric, glass, or pale objects need close review.
Busy room, hair, fur, or reflections Color alone rarely describes the subject boundary. A subject-aware editor or manual mask is usually safer.

Transparency is only useful after the edge has been checked against destination colors. A cutout that looks clean on a checkerboard may show a pale halo on a dark design, or a dark fringe on a white catalog page. Reviewing the same PNG against more than one matte catches those problems before the image is placed into final artwork.

The main limit is that color rules do not understand objects. They judge pixels by distance from a sampled color, brightness, connection to the border, and optional edge feathering. That makes them useful for simple assets, but it also makes manual review part of the workflow whenever the subject and backdrop are visually similar.

How to Use This Tool:

Start with the method that matches the backdrop evidence, then tune the mask until the preview, warnings, and coverage numbers tell the same story.

  1. Drop, paste, or browse one source image. The browser must be able to decode the file, and the file must be 40 MB or smaller. If an error appears, choose a smaller PNG, JPEG, WebP, GIF, SVG, AVIF, BMP, or another browser-readable image.
  2. Set Removal method. Choose Edge-sampled backdrop when the background touches the image border, Picked color key when a flat color should disappear everywhere, or Light studio backdrop for white or near-white product photos.
    Picked color key can remove matching colors inside the subject because it is not limited to border-connected pixels.
  3. For Picked color key, choose a Background color from the backdrop. Use Use edge sample only when the border represents the color you want to remove.
  4. Adjust Sensitivity in small steps. Raise it when paper, wall color, or shadows remain; lower it when the subject starts losing similar tones.
  5. Set Edge softness and Render ceiling. Use 0 px for crisp icons, about 2 to 4 px for many product photos, and a higher render ceiling when preserving more source detail matters.
    A lower render ceiling can speed up large images, but the exported PNG uses the rendered dimensions, not the original full pixel count.
  6. Open Advanced and switch Preview background between checkerboard, light, dark, and blue mattes when halos or semi-transparent edge residue could matter.
  7. Check Cutout Image, Removal Metrics, Mask Diagnostics, and Mask Coverage Chart. If the fit label says Review edge or Manual review, revise the method, color, or sensitivity before downloading the PNG.

Interpreting Results:

Removed pixels are fully transparent in the PNG. Softened edge counts pixels whose alpha was reduced near the removed background. Retained subject is the area kept opaque or mostly opaque after masking.

The fit label is a warning signal, not proof that the cutout is correct. A high fit can still be wrong when the subject shares color with the backdrop. A manual-review result usually means too little was removed, too much was removed, the border sample was mixed, or color keying may cut into the subject.

Output Cue What It Means What To Check
Very low removed percentage The mask found little background evidence. Raise Sensitivity or try a method that better matches the backdrop.
Very high removed percentage The mask may be cutting into the subject. Lower Sensitivity or choose a more specific Background color.
Large softened edge percentage Many pixels became partially transparent near the boundary. Inspect the cutout on dark and light mattes before using it in final artwork.
Mixed edge sample warning The border contains several colors, not one clean backdrop. Use color keying or prepare a cleaner source image.

Use the transparent PNG only after the visible edge, warnings, and coverage chart agree. When they disagree, trust the visual edge first and treat the metrics as troubleshooting clues.

Technical Details:

Color-based masking treats the source as a rectangular raster of red, green, blue, and alpha samples. The mask does not identify people, products, hair, or glass as semantic objects. It changes opacity by deciding which pixels are background, which pixels remain subject, and which pixels near a boundary receive partial alpha.

PNG is a useful target for this kind of result because it can store a full alpha channel. In a truecolor PNG with alpha, zero alpha means fully transparent, the maximum alpha value means fully opaque, and intermediate values can be composited over a new background.

Formula Core:

The edge-sampled and color-key paths use RGB distance from a sampled backdrop color. Sensitivity widens the tolerated distance.

d = (R-Rs) 2 + (G-Gs) 2 + (B-Bs) 2

Here R, G, and B are the current pixel channels, and the subscripted values are the sampled backdrop color. The current sensitivity number is converted into a tolerance with this rule:

T = S × 2.28

A 38% sensitivity setting therefore accepts pixels within about 86.6 RGB-distance units of the sampled backdrop color. Lower values protect similar subject colors, while higher values tolerate uneven paper, walls, and shadow variation.

For white or near-white studio backgrounds, brightness and color spread also matter. The luma estimate follows Rec. 709-style channel weights, and chroma is measured as the distance between the strongest and weakest channel.

Y = 0.2126 R + 0.7152 G + 0.0722 B
C = max ( R , G , B ) - min ( R , G , B )

In the light-backdrop path, bright low-chroma pixels connected to the border are background candidates. The cutoff becomes more permissive as tolerance rises: luma must be at least 252 - 0.34T, chroma must be no more than 8 + 0.32T, or the pixel must be close enough to white within 0.8T.

Rule Core:

Method Mask Rule Best Fit Failure Mode
Edge-sampled backdrop Averaged visible border color, then connected similar pixels are flood-filled inward from the edges. Clean product backdrops that touch the image border. Busy or multicolor borders make the backdrop estimate noisy.
Picked color key Every pixel is compared with the chosen color, even away from the edge. Flat-color assets, green-screen style images, and logos. Matching colors inside the subject can become transparent.
Light studio backdrop Connected pixels pass when they are bright, low in chroma, or close enough to white. White or near-white studio photos with simple shadows. Pale subjects, glass, and gray objects can be over-removed.

Edge softness applies after the background mask is found. A pixel that is next to a removed pixel can receive reduced alpha when it is still close to the backdrop color. That feathering helps jagged product edges, but too much softness can make a solid subject look translucent.

Large sources are proportionally scaled before masking when their pixel count exceeds the selected render ceiling. The scale uses the square root of the ceiling divided by source pixels, which preserves aspect ratio.

scale = render ceiling source pixels
Diagnostic Trigger Threshold Interpretation
Very little removed < 1% removed pixels The selected method or tolerance probably missed the backdrop.
Most pixels removed > 92% removed pixels The mask is likely eating into the subject or choosing the wrong key color.
Mixed border sample Diversity > 38 The image edge is not a clean representation of the background.
Large feathered edge > 16% softened pixels Partial transparency is widespread enough to require visual checking.

GIF input is exported as one rendered frame, and SVG input is rasterized before masking. The output is a transparent PNG bitmap, so animation, vector editability, original metadata, and source format-specific features are not preserved.

Privacy Notes:

Image decoding, mask creation, preview, metrics, and PNG export happen in the browser tab. Local image pixels are not uploaded for background removal. Use a smaller file or lower Render ceiling if a very large image makes the tab slow, and remember that the lower ceiling also reduces the exported pixel dimensions.

Worked Examples:

A 2400 x 1600 product photo on a pale gray sweep is a typical Edge-sampled backdrop case. With Sensitivity near 38%, Edge softness at 3 px, and Render ceiling set to 3 MP, Removal Metrics should show a downscaled render near the ceiling, a meaningful Removed pixels share, and a smaller Softened edge share. A High fit label is still worth checking on a dark matte when the product edge is light gray.

A 1600 x 900 logo on a green field is better suited to Picked color key. Choose the green Background color, start with moderate Sensitivity, and keep Edge softness at 0 px if the logo has crisp vector-like edges. If Retained subject shrinks or a warning mentions matching subject tones, lower sensitivity or choose a color farther from the logo strokes.

A 5000 x 4000 JPEG at the 6 MP Render ceiling will be reduced to roughly 2739 x 2191 before masking. That is expected because the ceiling caps browser-side pixel work. Mask Diagnostics should include the downscale note, and the exported PNG will use the rendered dimensions rather than the full 20 MP source size.

A troubleshooting pass often starts with the warning Very little background was removed. If the source is a white studio photo, try Light studio backdrop; if it is a clean flat color, switch to Picked color key; if the border is the right backdrop but shadows remain, raise Sensitivity a few points and recheck the Mask Coverage Chart.

FAQ:

Is this AI background removal?

No. It uses browser-local color, edge, brightness, and feathering rules. Complex subjects such as hair, fur, glass, smoke, and cluttered scenes need manual review or a subject-aware editor.

Why does the cutout still have a halo?

A halo usually means the backdrop is uneven, sensitivity is too low, or edge softness is not matching the subject boundary. Switch the Preview background matte and adjust Sensitivity in small steps.

Why did part of the subject disappear?

The chosen background color or tolerance is too close to a subject color. Lower Sensitivity, choose a different Background color, or switch away from Picked color key.

Why is the output smaller than my source image?

The Render ceiling caps browser-side pixel work. Images larger than the selected ceiling are proportionally downscaled before the PNG is generated.

Are my images uploaded?

No upload is used for background removal. The image is decoded, masked, previewed, and exported inside the browser tab.

Can it preserve animation or vector editability?

No. GIF input exports as one rendered frame, and SVG input is rasterized before masking. The final file is a transparent PNG bitmap.

Glossary:

Alpha
The transparency value stored for a pixel, from fully transparent to fully opaque.
Mask
The pixel-by-pixel decision that marks background, retained subject, and softened edge areas.
Color key
A selected backdrop color used as the reference for removing matching pixels.
Edge sample
The average color estimated from visible pixels around the image border.
Feathering
Partial transparency applied near a cutout boundary to soften jagged edges.
Luma
A weighted brightness estimate based on red, green, and blue channel values.
Chroma
The spread between the strongest and weakest color channels, used here to identify low-color white or gray backdrop pixels.

References: