YouTube thumbnail size in 2026

What YouTube's own documentation says now

YouTube's help page for custom thumbnails currently gives a recommended resolution of 3840 x 2160 pixels for standard videos, in a 16:9 aspect ratio. The minimum accepted width is 640 pixels. Accepted formats are JPG and PNG. The file size limit depends on where you upload from: 2 MB when uploading from a mobile device, and up to 50 MB from desktop.

For Shorts the numbers flip: 2160 x 3840 pixels in a 9:16 ratio, with a minimum height of 640 pixels. If you publish both formats, you are working with two different canvases, not one cropped two ways.

Why the recommendation went up

The 1280 x 720 figure dates from a time when almost nobody watched YouTube on a screen denser than a 1080p monitor. Today a large share of viewing happens on phones with pixel densities of two or three times that, and on televisions running at 4K. When YouTube displays your thumbnail on a 4K television home screen, a 1280-pixel-wide image is being stretched to three times its native width.

Raising the recommendation does not change what is accepted. A 1280 x 720 thumbnail still uploads and still displays. What changes is how it looks on the largest and densest screens, which is exactly where the competition for a click is most visual.

So what should you actually export?

If you only ever export one size, 1920 x 1080 is the pragmatic sweet spot. It doubles the linear resolution of the old standard, it is a size every design tool handles without straining, and it stays comfortably under the mobile upload limit as a JPG.

If your thumbnail contains a photograph and you upload from desktop, go to 3840 x 2160 and let YouTube do the downscaling. Their downscaler is better than yours, and you keep the detail for the screens that can show it.

Where 1280 x 720 still makes sense is when your design is flat colour and heavy type, and you need a small PNG for a mobile upload. At that point the extra pixels buy you very little, because there is no fine detail to preserve.

Enlarging what you already have is not the same as exporting bigger

Once you know the recommendation is 3840 x 2160, the obvious move is to open the thumbnail you already made and type the new numbers into the resize dialogue. Nothing stops you, and the file that comes out really is 3840 pixels wide. It simply does not contain any more information than the one you started with.

Enlarging spreads the pixels you captured across a bigger grid and fills the gaps by averaging neighbours. Detail that was never recorded cannot be reconstructed by arithmetic. What you get is the same picture, softer, in a heavier file — and often slightly worse than that, because interpolation smears edges that were previously clean, and sharpening afterwards adds a bright fringe along them rather than restoring anything.

So the resolution decision is made when you gather your material, not when you export. Pull the still from your editing timeline at full resolution instead of screenshotting the preview window. Use the original camera file rather than the copy that came back through a messaging app. Use the image from the press kit rather than the one you saved off a web page. Whatever the largest version you can lay hands on is, start from that.

If your only usable source genuinely is small, design around it instead of stretching it. Text, shapes and flat colour are drawn at whatever size you export, so they stay sharp regardless; it is only the photographic element that has a ceiling. Set that element smaller within the frame, or crop tighter into the part of it you actually need, and export the whole thing at 1920 x 1080. A canvas four times larger than anything you can put inside it buys you nothing.

The upscalers that promise to add detail are doing something more interesting than averaging, but they are inventing plausible pixels rather than recovering the ones that were there. On texture such as foliage or fabric the result can be convincing. On faces and on lettering it tends to come out sharp and subtly wrong, which in a thumbnail is more distracting than softness would have been.

The aspect ratio is the part you cannot compromise on

Resolution is a quality question. Aspect ratio is a cropping question, and cropping is what actually destroys thumbnails. Anything that is not 16:9 gets fitted into a 16:9 box, which means either bars down the sides or a centre crop that eats your text.

This catches people who design a square image for Instagram and then reuse it on YouTube. The square gets cropped top and bottom, and whatever was in those strips disappears. If you want one design across platforms, build it at 16:9 and crop outward to square, never the reverse.

The file size limit is about where you upload from

The 2 MB figure that everybody quotes applies to uploading from the YouTube mobile app. From a desktop browser the ceiling is far higher. This matters because a 3840 x 2160 PNG of a photographic thumbnail will very often exceed 2 MB.

The practical rule: if you upload from your phone, export JPG and keep an eye on the file size. If you upload from a computer, you have room to use PNG at full resolution and let YouTube handle the compression.

Check it at the size it will be seen, before you upload

Every design tool shows you your thumbnail far larger than any viewer will ever see it. The check that catches most problems is to look at it at about 210 pixels wide, which is roughly what a phone gives a thumbnail in the feed, and at around 168 pixels, the width of the suggested-video column on a desktop watch page. If it still reads at those sizes, the rest is detail.

You do not need a special tool for this. Zoom the canvas out until the image is about the width of your thumb, or open the exported file and shrink the window. What you are looking for is whether the main word is still a word and the face is still a face, not whether the design is attractive.

Then take the colour out. A quick desaturation over the whole design tells you whether your subject separates from its background by tone or only by hue. Colour that looks strong against the white background of an editor often turns out to be the same brightness as the thing behind it, and at small sizes tone is what the eye uses to separate shapes. If everything collapses into a similar grey, the fix is to darken or lighten something, not to make the colours louder.

Finally, squint at it, or apply a heavy blur and look at what survives. You should be able to say what the thumbnail is about from the blurred version alone. If two or three elements survive with equal weight, the design has no hierarchy, and at feed size a viewer scrolling past gives it rather less attention than your squint did.

What YouTube does to your file after upload

Whatever you send is re-encoded. YouTube generates its own set of derivatives at several sizes and serves whichever fits the surface. That second pass of compression is why a heavily compressed JPG looks worse on YouTube than it did on your desktop: you are stacking two lossy passes on top of each other.

The defence is simple. Give the encoder the cleanest possible input. A large, lightly compressed source survives re-encoding much better than a small, heavily compressed one, even when the final displayed size is identical.