Guide
Create Transparent PNG
Short answer: transparency is not produced by a button, and there is no single “make it transparent” operation. It gets into a file from exactly three places: a colour decision (you remove a background), a drawn shape (you define an outline that everything outside of is empty), or an image that already carries it (you paste something that was cut out somewhere else). Each of the three fails in its own recognisable way, and only one thing goes wrong when you combine them — which is the part almost nobody writes down: alpha multiplies when you stack it. Two half-transparent things over each other do not give you half; they give you a quarter.
Where the empty background actually comes from
| Source | What you do to produce it | How it typically fails | Right for |
|---|---|---|---|
| A colour decision | Nominate a background colour and remove everything close to it | Fringing and spill at the outline; holes where the subject shares the colour | Photographs and screenshots of real things |
| A drawn shape | Define an outline — a path, a mask, a selection — and treat everything outside it as empty | Hard corners and thin strokes that break up when the shape is rasterised | Logos, icons, lettering — anything with a defined boundary |
| An image that already has it | Place a PNG or WebP that was cut out earlier | Nothing, until you stack two semi-transparent things — then the trap below | Assembling a composition from parts |
1. From a colour
This is the one every guide covers, so keep it short. The tool asks, per pixel, how far the colour is from the one you nominated; far means keep, close means delete. Everything that can go wrong is at the outline or on the subject itself, and both are fixable with the right control rather than more tolerance. The four decisions are: sample a flat patch, raise tolerance only until the colour is gone, rebuild the edge, cap the spill.
Its characteristic weakness is that it only knows colour. Anything in the picture that happens to be the background colour goes with it, and anything that carries a tint of it stays.
2. From a shape
Here you are not measuring anything — you are declaring. Everything inside the path is a subject; everything outside is empty. That makes it exact and repeatable, which is why logos and icons are almost always cut this way rather than by colour.
Its failure mode is geometric rather than chromatic. A path is smooth at any scale while you edit it, but the exported PNG is a grid of pixels, and the moment it is rasterised every curve and every sharp corner becomes a decision about which pixels to include. Thin strokes are the first casualty: a one-pixel line at a low export size will break into a dotted line or disappear. The practical rules are to export large and scale down, and to avoid strokes thinner than about two pixels at the final size.
3. From an image
The third source is the one people use without noticing: you paste a picture that is already transparent. Nothing is measured and nothing is drawn — the alpha simply arrives with the file, which makes this by far the cleanest way to build a complex transparent PNG. It is also the only one of the three that can silently damage itself when you combine pieces.
The trap: alpha multiplies when you stack
This is the part worth the read. Alpha is not a label that says “transparent”; it is a number between 0 and 1, and when two layers are composited the numbers multiply.
- Half over half is a quarter. A pixel at 50% opacity placed on another pixel at 50% opacity ends up at 25%. Not 50, not the average — a quarter.
- It compounds fast. Three such layers give 12.5%. A soft-edged logo placed on a translucent panel and then exported inside another translucent container can end up barely visible, and nothing on screen explains why.
- Edges are already fractional. Remember that a cut-out’s outline is itself a blend — those pixels are, say, 0.4. Stack a second soft layer over them and the outline thins again, in the same way it does when you re-key a finished cut-out.
What to do about it is simple once you know: decide each piece’s opacity once, at the size it will be seen, and avoid putting semi-transparent things inside other semi-transparent things. If a composition needs a translucent panel, make the panel opaque in the file and let whatever displays it supply the transparency — one layer of alpha, not two. And if you are exporting a final image meant to be partly see-through, flatten deliberately at the end so the number you chose is the number in the file.
Which one to use
- Something real, photographed: colour. There is no path to draw around hair, foliage or a crumpled cloth, and no existing alpha to inherit.
- Something with a defined boundary — a logo, an icon, lettering, a diagram: a shape. It is exact, it scales, and it does not care what colour anything is.
- Anything already cut out: place it and stop. Re-cutting a finished piece is the one thing that reliably makes it worse.
- A composition: cut the photographic parts with colour, draw the geometric parts as shapes, then assemble from those clean pieces — and keep the layer count of semi-transparent things down to one.
In practice most real files use two of the three, and that is fine as long as you know which piece came from where. What goes wrong is using one mechanism to try to fix the problems of another: re-keying a drawn logo to clean up its corners, or drawing a path around a person’s hair.
Testing the file you created
Three checks, all free, in this order:
- Put it on a dark background. The empty area should show the dark. If a rectangle appears, the transparency never made it into the file — go back to the export settings rather than the cut-out.
- Look at a corner at 100%. Corners are where a colour key leaves a fringe and where a path leaves a stair-step; they are the first place either mechanism shows itself.
- Place it where it is actually going. A cut-out judged on a neutral grey can look perfect and then read as a sticker on a dark page, because the edge that worked against grey does not work against black.
This site’s tool, honestly
The remover here implements exactly one of the three mechanisms: the colour one. It measures how far each pixel’s colour sits from a colour you nominate, which makes it the right instrument for photographs and the wrong one for logos you could simply draw. It does not create canvases, does not draw paths, and cannot tell what a subject is — so if your file is a geometric shape or a stack of existing cut-outs, you are already holding the better tool for the job.
What it is good at is turning a photo shot against one flat colour into a clean transparent PNG, in the browser, without uploading anything. Open the green screen remover when that is the piece you need.
If the removal looked right and the exported file still has a background, that is an export problem rather than a removal one: this page separates the three kinds of file that all get called a transparent PNG.
Questions
How do I create a transparent PNG?
The transparency has to come from somewhere. Either you remove a background colour from a picture, or you draw an outline around a shape, or you place an image that already carries an empty background — then you export as PNG with nothing flattened. Pick the mechanism that matches what you are holding.
Why does my transparent PNG look faded or ghostly?
The most likely cause is stacked transparency: alpha values multiply when layers are composited, so two 50% layers become 25%. If a soft-edged element sits on a translucent panel, both numbers apply at once and the result is much weaker than either one alone.
Should I remove a background or draw a shape?
Remove for anything photographed, because you cannot trace around hair or a crumpled surface and get it right. Draw for anything with a defined boundary — a logo, an icon, lettering — because a path is exact and does not depend on the colours in the picture.
Why do thin lines disappear when I export a logo?
Because the path is smooth but the PNG is a grid. A stroke thinner than about two pixels at the export size has to be rounded to whole pixels, so it becomes dotted or vanishes. Export larger and scale down, and avoid strokes that thin at the final size.
Can I mix all three on one file?
Yes, and most real compositions do: photographic parts cut by colour, geometric parts drawn as shapes, everything assembled from clean pieces. The only rule that matters when you mix them is to keep the number of semi-transparent layers to one.
Does the file type matter for this?
Completely. Transparency is stored in a channel, and only some formats have one. PNG and WebP carry it; JPEG cannot, so any JPEG export fills the empty background again no matter how clean the cut-out was.
Can I do this on a phone?
The colour mechanism, yes — the remover on this site runs in the browser, so the image is processed on your own device and never uploaded. Drawing precise shapes and assembling layers is still much easier with a larger screen and a pointer.
Open the green screen remover — free, no account, and it writes a PNG with a real empty background at the size you loaded. If you are still deciding what kind of file you have in the first place, this page separates creating an empty canvas from removing what is already there.