Guide

Six ways to fill an edge that was never photographed

None of these knows what was outside your frame. All six are arithmetic on pixels you already have — which is a limitation and, on the right subject, exactly the right answer.

What every fill on this page has in common

Each one takes the photograph you brought and produces new pixels from it by a rule. Reflect it, smear it, soften it, average it, sweep between two averages, or decline to put anything there at all. That is the whole set, and the honesty of the whole site rests on saying so plainly: nothing here is a guess at the room you were standing in, and nothing here has ever seen a photograph other than yours.

Which is a real limit, and also the reason these fills are fast and predictable. A rule applied to your own pixels takes a few milliseconds, runs the same on every device, and fails in ways you can see coming. The rest of this page is about seeing them coming.

The six, one at a time

Mirror — the reflection

The strip nearest each new edge is flipped and drawn outward, alternating direction if the space is wider than the photograph. Values match perfectly at the join, because one side is literally the other read backwards, and on the right subject the seam genuinely disappears.

It flatters anything without direction or landmarks: foliage, gravel, sand, water, knitted fabric, a crowd at distance, a textured wall. It ruins anything the eye can name. A face touching the edge comes back as a twin. A door frame becomes a double door. A line of text becomes its own reverse. And past two reflections the repeat itself is the giveaway, which is what the fade-to-blur option is for.

Edge stretch — the smear

The outermost single column or row is drawn outward across the whole new region, so every added pixel takes its colour from the one photographed pixel directly beside it. Corners take theirs from the corner pixel, because when both axes have grown that is the only answer with any claim to honesty.

It is unbeatable on a gradient with nothing in it: a clear sky, a studio backdrop, a plain painted wall, a sunset. It is the worst of the six on anything else, because a structure of any kind becomes a horizontal or vertical streak that reads instantly as a software mistake. Use it when the edge of the photo is boring, and only then.

Blurred cover — the scenery

A copy of your photograph is enlarged until it covers the entire target canvas, then blurred until no detail survives, then optionally taken up or down in tone. Your original is laid over the middle of that, at its own size and full sharpness. The backdrop is scenery; the picture is the picture.

It works on nearly everything, which is why it is the default, and it works best on busy photographs where a strong mix of colours makes an interesting field. It has two weaknesses. On a very plain photograph the backdrop is a flat wash and you may as well have chosen a colour. And on a portrait where a face sits near the edge, the enlarged copy can put a soft ghost of that face behind the sharp one — nudge the photograph away from the centre and the ghost moves with it.

One colour — the flat field

A single value across the whole new region. Four swatches, a hex field, and the option that matters most: the average colour of your own photograph's outer border, which is nearly always a better answer than anything you would have picked deliberately.

It is the right choice for documents, print margins, and anything with a hard-edged subject that would look absurd reflected. It is also the right choice when the picture needs to look framed rather than extended — a photograph on white reads as a photograph on white, which is a legible thing to be. Its failure mode is bland, never wrong.

Gradient — the sweep

Two colours, taken by default from opposite edges of your own photograph, swept across the whole canvas at an angle you choose. It is a flat fill with somewhere to go, and it suits photographs that are already a gradient themselves: skies, water, anything shot against a lit backdrop that falls off across the frame.

Set the angle to match the direction the picture already travels in and the join stops announcing itself. Set it across that direction and you have drawn attention to exactly the seam you were trying to hide, which is the only real way to get this one wrong.

Nothing at all — the empty region

The sixth is the one that keeps getting left off the list, including on this page until recently: the new region is not filled. No reflection, no smear, no wash — the pixels are written with an alpha of zero and the photograph sits in the middle of genuine emptiness. It is a real answer to the question, and on the way into a layout it is frequently the right one, because whatever the picture is placed on becomes the fill.

It has one hard requirement and the tool states it before you save: a PNG or a WebP can record an empty pixel and a JPG cannot. Choose this fill and write a JPG and you get white, because white is what the encoder has to put there. The panel says so beside the control rather than letting the file be the first place you find out.

A table to decide by

The fill to reach for first, by what is at the edge of your photograph
At the edge of the frame	Reach for	Avoid
Clear sky, plain wall, studio paper	Edge stretch	Mirror
Leaves, gravel, water, fabric	Mirror	Edge stretch
A person, a face, a hand	Blurred cover	Mirror
Architecture, tiles, lettering	One colour	Mirror and stretch both
A sunset or a graded backdrop	Gradient	One colour
Going into a document or a print	One colour	Blurred cover
Going into a layout that has its own background	Nothing at all	Anything that fills

Each row is a starting point and not a verdict. The rail names every fill as you pass it and tells you which ones the shape you have chosen rules out; the picture on screen is the only thing that settles it.

Since this was written the tool has gained a seventh arithmetic fill, smart extend, which sits between the mirror and the stretch: the texture right at the seam is carried across, and then the edge’s own colours run outward along the lines that meet the frame, softening the further they travel. It is the one to try first on a horizon or a graded sky that a plain stretch would streak, and it fails on a face at the edge in the same way the other six do.

The awkward questions

Is there a fill that is right more often than the others?

The blurred cover, by a wide margin, and not because it is cleverer — because it has no structure in it. Every other fill puts recognisable shapes beside your photograph and therefore invites the viewer to check whether those shapes make sense. A blur makes no claim at all, so there is nothing to catch on. It is the one to reach for when you have not got time to look closely at the result.

Can I combine two of them?

One combination is built in and it is the useful one: a mirror that gives you a single reflection and then dissolves into a blurred backdrop. It exists because the first reflection is where mirroring is genuinely convincing and the third is where it starts looking like wallpaper. Beyond that, no — mixing fills per side would be four times the controls for a case that comes up rarely.

Why does the mirror need a softening slider if the colours match exactly?

Because what jumps at a mirrored join is not colour, it is direction. A branch travelling up and to the right meets a branch travelling up and to the left, and the eye is extremely good at spotting that corner even when every value on both sides is identical. Blurring a narrow band across the join turns the corner into a curve. Your photograph is then drawn back over the top at full sharpness, so the softening lands only on the invented side.

My photo has a transparent background and mirror does nothing.

It is doing exactly what it was asked to — reflecting the strip nearest the edge, which on your file is empty. Copying nothing outward gives you more nothing. The tool notices transparency at the edges and says so in the panel; on a file like that, one flat colour, a gradient, or leaving the new region empty are the three answers that work, and the last one needs a PNG or a WebP to survive.