Resizing an Image for Each Network Without Distorting It

Never stretched, always centred

The first question about any resizer is whether it distorts. This one does not, and the reason is one line: it computes a single scale factor and applies it to both axes. Fill takes the larger of the two ratios so the frame is covered; Fit takes the smaller so the whole picture is inside. Either way width and height are multiplied by the same number, so faces do not stretch and circles do not become ovals. That is worth confirming rather than assuming, because a resizer that simply forced the source into the target rectangle would be indistinguishable in a thumbnail and obvious the moment anyone looked.

The cost is paid elsewhere. In Fill mode the overflow has to go somewhere, and the code puts the image at half the difference on each axis — dead centre, every time, with no focal point, no rule-of-thirds anchor, and no way to drag the frame. Whether that is fine or ruinous depends entirely on how far the target ratio is from the source. Feed a 1000 by 1500 portrait through the presets and the damage is measurable: the 4:5 portrait target takes 8.3 percent off the top and the same off the bottom, the square takes 16.7 percent each way, the 1.91:1 link preview 32.5 percent, the 3:1 banner 38.9 percent, the 4:1 cover 41.7 percent. Going the other way, a 1920 by 1080 landscape into the 9:16 story preset loses 34.2 percent off each side.

So the failure people actually hit is real and predictable. A portrait photograph where the subject's head sits in the top third, pushed through a banner-shaped preset, comes back as a strip of torso. The tool will not warn you, and the thumbnail it shows in the results grid is displayed with object-contain, so the preview is honest — but you have to look at it. Look at every result before downloading the batch, not just the first one.

What to do when the centre is the wrong place

Since the frame cannot be moved, move the picture. Crop the source yourself first, to something close to the widest target ratio you need, putting the subject where you want it. Then feed that crop to the resizer: the remaining ratio differences are small, so the centre crop removes a few percent instead of forty. In practice, for a set that includes 1:1, 4:5 and 1.91:1, cropping the original to roughly 4:5 with the subject centred vertically gets every output within about a tenth of the frame.

Fit is the other escape, and it is the right answer more often than its reputation suggests. It keeps the whole picture and pads the remainder with a colour you pick, which for a product shot, a chart, a logo lockup or a screenshot is exactly what you want: nothing at the edges is load-bearing, and losing none of it beats framing it well. Choose the padding colour to match the background of the image itself and the padding disappears visually. Where Fit does look wrong is on a photograph going into a very wide frame — the same 1000 by 1500 portrait into the 1584 by 396 cover occupies 264 pixels of width, one sixth, with the rest a flat block of colour. At that point the honest answer is a different photograph, not a different setting.

Upscaling is silent here, and it never adds detail

Neither mode clamps the scale factor at one. If the source is smaller than the target, the picture is enlarged, and there is no message about it anywhere in the interface. Measured: a 400 by 400 source into the 1080 by 1080 square is drawn at 2.7 times, into the 1080 by 1350 portrait at 3.375, and into the 1080 by 1920 story at 4.8. Even a respectable 1000 by 1000 source is drawn at 1.584 times for the 1584 by 396 cover. The output file will carry the full target dimensions and will look, in a file listing, exactly like a properly sized export.

It will not look like one on screen. Enlargement invents pixels by interpolating between the ones that exist; the browser's smoothing makes the result soft rather than blocky, which is prettier and no more informative. There is no detail in an image that the sensor did not record, and no amount of scaling puts it there. If your source is too small, the fixes are, in order: find the original before it was resized for something else, re-export it from whatever produced it, shoot or render it again at the target size, or accept the smaller placement. Every platform in the table below states a minimum and enlarges below it themselves — Instagram enlarges anything narrower than 320 pixels to 320, and its own guidance is to upload at least 1080 wide.

The dimensions, with the date each one was checked

A table of social dimensions with no date on it is worse than none, because it will be wrong and there is no way for the reader to tell when it stopped being right. Every figure below was read on the platform's own help or developer page on 25 August 2026, and the surprise is at the top of the list. Instagram's page on photo resolution now gives the supported feed range as 1.91:1 to 3:4, which at 1080 wide means a height between 566 and 1440. Not 1350. The 4:5 figure that every guide repeats — and that this tool's own portrait preset uses — is inside the supported range but is no longer the tallest thing the feed accepts. Search engines were still serving the older 4:5 wording from their index on the same day the live page said 3:4, which is exactly how a stale number outlives the page it came from.

The second surprise is YouTube. Its help page gives 3840 by 2160 as the thumbnail resolution, with a minimum width of 640, a 16:9 ratio and a file cap of 50 MB on desktop but only 2 MB on mobile. The 1280 by 720 that this tool offers, and that the whole internet quotes, is a valid 16:9 size well above the minimum, but it is not what the platform publishes. Three figures I could not verify from a primary source at all, and so left out of the table: the X in-feed image and header sizes, because help.x.com and business.x.com both refuse automated reads; TikTok's video resolution, because the ads help article that documented it no longer exists; and the Facebook page cover, because the help page renders nothing to read. Those presets may well be right — I simply cannot show you where they come from, and a figure I cannot source does not belong in a table that claims to be checked.

Safe area, and why 1:1 is the robust default

One image is rarely shown once. A post image appears in the feed at the ratio you uploaded, in the profile grid as a square, in a shared link preview at roughly 1.91:1, and in a notification as a small thumbnail — each of those a different crop of the same file, performed by someone else's code with no reference to your intentions. That is the safe-area problem, and the only defence is to leave the edges empty. Keep text, faces and logos inside the central square of whatever you upload, and treat the outer band as decoration that may or may not survive. Anything you would be unhappy to lose does not belong in the top or bottom eighth.

That is also the case for a square export as the default when you do not know where the file will end up, and the numbers back it. Run a 1000 by 1000 square source through every preset and the worst vertical loss is 37.5 percent each way, into the 4:1 cover; a 4:5 portrait source into the same target loses 40 percent each way, and a 16:9 landscape into the 9:16 story loses 34.2 percent off each side while the square loses 21.9. A square is never the farthest thing from any target, which is the whole property you want from a default. It is not the best export for any single placement — 4:5 takes more room in an Instagram feed, 1.91:1 is what a link preview actually renders — but it is the one that degrades least when it lands somewhere you did not plan for.

Two neighbouring tools finish the job. The carousel maker turns lines of text into 1080-pixel slides and opens in portrait at 1080 by 1350, with square and story also available, exporting PNG — useful precisely because text rendered at the export size never goes through a second resampling. And the post preview tool answers the other half of the question: it separates the fold, meaning how much of your caption is shown before the reader has to tap, from the hard limit the composer refuses to exceed. It encodes 125 characters before the fold and a 2200-character limit for Instagram, 140 and 3000 for LinkedIn, and for X no fold at all with a 280-character limit, since nothing is hidden there. Everything in all three runs on a canvas in your own browser; no file is uploaded anywhere.

Frequently asked questions

Does this tool crop, letterbox or stretch?

It crops or letterboxes, never stretches. Fill applies one scale factor large enough to cover the frame and lets the excess fall outside it, cropping equally from the two opposite edges. Fit applies one scale factor small enough for the whole picture to sit inside, and pads the remainder with the background colour you choose. Both preserve the aspect ratio exactly, because the same factor is applied to width and height. The crop position is fixed at the centre and cannot be changed, so if the subject is off-centre, crop the source yourself before you feed it in.

Which single size should I export if I only make one?

1080 by 1080. It is not optimal anywhere and it is never terrible anywhere, which is what a default is for. Measured across the presets, a square source loses at most 37.5 percent per side into the widest cover and 21.9 percent per side into a 9:16 story; a 4:5 portrait source loses 40 percent into that same cover and a 16:9 landscape loses 34.2 percent into the story. If you know the destination, export for it: 1200 by 630 for a link preview, 1080 by 1350 for an Instagram feed post, 1000 by 1500 for a Pin.

My source photo is smaller than the target. What happens?

It is enlarged, silently, with no upper bound. Measured: 400 by 400 into the story preset is drawn at 4.8 times, and the resulting file carries the full 1080 by 1920 dimensions as if it had been made at that size. Nothing in the interface warns you. Since enlargement interpolates between existing pixels and cannot invent detail, the result is soft rather than sharp. Find a larger original, re-export from the source application, or accept a smaller placement — those are the only real fixes.

Should I export JPEG or PNG?

JPEG for photographs, PNG for anything with flat colour, sharp edges or text — a chart, a logo, a screenshot, a carousel slide. The reason is that JPEG discards high-frequency detail, which is imperceptible in foliage and produces visible fringing around a letter. The tool offers a quality slider on the JPEG side, which is worth using rather than leaving at whatever it defaults to. Note that the copy-to-clipboard button always re-encodes as PNG regardless of your choice, because browsers only accept PNG on the clipboard; the download keeps the format you picked.

Why are some networks missing from your table?

Because I could not read the figure on the platform's own page, and a number I cannot source does not belong in a table that claims to be checked. On 25 August 2026, help.x.com refused automated reads and business.x.com returned a payment-required response, so both the X in-feed size and the header size are absent. TikTok's video specification article in its ads help centre no longer exists. Facebook's page on cover photo dimensions rendered no readable body. The presets those figures come from may still be correct — treat them as sensible conventions rather than as published requirements, and check the platform yourself before a launch depends on it.