YouTube thumbnail specs.

The spec sheet

The numbers themselves are short work. YouTube’s help page asks for an aspect ratio of 16:9 for videos, recommends a resolution of 3840×2160 pixels (16:9) for videos; 2160×3840 for Shorts, and sets a minimum of 640 pixels wide (videos); 640 pixels tall for Shorts. The file-size limit depends on the device you upload from: 2 MB when uploading from mobile; 50 MB when uploading from desktop (video thumbnails). The accepted formats are stated only as “Image formats such as JPG or PNG”. Everything else about thumbnail design is judgment, but these five values are the gate your file has to pass first.

Every figure in the table was read on 2 October 2026from YouTube’s own article on custom thumbnails (source). Verify against the platform’s own documentation before a critical print run or launch.

Two things follow for the thumbnail tool. Its canvas is fixed at 1280×720 pixels — the 16:9 ratio, 640pixels wider than the minimum — so its export passes the gate but is no longer the recommended resolution, and this guide no longer calls it that. And the smaller file-size figure, the mobile one, is the one an export must stay under to work from either device.

The 168-pixel test

Here is the part the spec sheet hides: viewers mostly meet a thumbnail at reduced sizes, not at 1280×720. On the site’s own wording, kept from August 2026 because no official YouTube page states render sizes, it often renders around 360 pixels wide in search results and shrinks to roughly 168pixels in the suggested-videos sidebar. Because that is the smallest size the design has to survive, the sidebar render is the binding constraint — and the real test of the design.

A thumbnail that looks rich and detailed at full size can collapse into mud at 168 pixels. Fine textures disappear, thin type becomes unreadable, and a busy composition turns into visual static. The practical habit is to design at 1280×720 but judge at sidebar size: shrink the canvas down, step back, and ask whether a stranger could tell what the video is about at a glance. If the answer requires squinting, simplify.

The shrink, in numbers

A 168-pixel render of a 1280-pixel canvas is 13% of its width, so every pixel the viewer sees stands for about 7.6canvas pixels, and type shrinks by the same factor: the Studio’s default overlay title of 88 pixels becomes roughly 11.6pixels tall in the sidebar, and the smallest overlay title the Studio’s normaliser lets you set, 48 pixels, roughly 6.3. The renderer may fit a long title smaller still, down to a floor of its own that sits below the setting’s. Hence the headline rule below: words, not characters.

The canvas is a screen asset, not a print one: by the QR engine’s print-size arithmetic at 300 DPI, 1280 pixels is 4.27 inches (108.4 mm) and 720 pixels is 2.40 inches (61.0mm). Print work scales the Studio’s SVG QR code instead; see the decode-margin guide.

Leave the bottom-right corner alone

YouTube overlays the video duration badge on the bottom-right corner of every thumbnail. Anything you place there — a logo, a word, the punchline of your headline — gets covered by a dark timestamp box wherever the thumbnail appears in a feed. Treat that corner as reserved space and keep critical elements out of it. The safest layout puts the headline in the upper half or along the left side, where nothing competes with it.

The badge and the two render widths above are the site’s own wording, kept from August 2026because no official YouTube page states them; the tool treats them as layout rules regardless, and with its safe-zone checkbox on the preview draws three dashed marks — the duration safe zone in that corner, a hover-controls region top-right, a progress strip along the bottom edge — that the export leaves out.

One subject, a short headline

The compositions that survive shrinking share two habits. First, they commit to a single subject — one face, one product, one object — that fills a large share of the frame. Two subjects halve each other’s size; three make everything anonymous. Second, they keep on-image text brutally short. A handful of large words gives the viewer a fair chance to read it at sidebar size; a full sentence gives them almost no chance to. The title below the thumbnail already carries the sentence — the image only needs to carry the hook.

• Pick one focal subject and let it dominate the frame.
• Limit on-image text to a few large words, not a caption.
• Keep strong contrast between text and background — the WCAG AA threshold of 4.5:1 for text is a sensible floor even though thumbnails are images, because a ratio that fails it at full size fails harder at 168 pixels.
• Avoid tiny logos and watermarks that read as noise when scaled down.

What the tool renders

The thumbnail maker draws on one fixed canvas of 1280×720 pixels and offers 5 templates (Editorial, Tutorial, Interview, Minimal, Before / after), each a composition rule rather than a decoration. A photo is drawn to cover the canvas, a directional shade is laid over the text side, and a darker protective scrim is laid under the text before any type is drawn. After export the file is decoded again and its size compared with the canvas; a mismatch is an error, not a warning — the verification the tool page calls mandatory.

A YouTube URL is reduced to one 11-character video id before anything is fetched: on fictional inputs, https://www.youtube.com/watch?v=pr0moLaunch&t=42s and https://youtube.com/shorts/pr0moLaunch both yield pr0moLaunch, while a non-YouTube host is refused with “This is not a supported YouTube URL.” The maker then fetches the platform’s own thumbnail images in order, from Maximum resolution (maxresdefault.jpg, at least 1000 pixels wide) down to Medium quality (at least 300), decodes each in the browser and keeps the first that clears both its width floor and the small height floor the engine keeps; if none does, it says so and waits for an upload.

The headline is fitted, never clipped

The Studio’s text-fitting function lays the title into the box the template leaves for it: from the preferred size it steps down one pixel at a time, wrapping the words at each size, until the lines fit the box’s height; only if they still do not fit at the minimum does it keep the lines that fit, end the last with an ellipsis and report truncation. Two different rules carry a size here, and they are easy to confuse. The Studio’s normaliser clamps the overlay title you set to between 48 and 120pixels, and the Studio hands that setting to the renderer as the preferred size — the default is 88. The thumbnail tool passes no size, so there the renderer starts from a preferred size of its own per template, smaller for Minimal. In both cases the floor the fitter steps down to is the renderer’s own constant, lower than the normaliser’s floor, with a line spacing of its own tighter than the fitter’s default; neither is restated here. Worked on fixed fictional inputs — the Studio default of 88 pixels as the preferred size, an example floor of 44pixels (half of it), the fitter’s default line spacing, and a flat measurement model of 0.56 em per character — in a box 768×432 pixels:

• Three ways to light a small kitchen fits at 88 pixels on 3 lines: “Three ways to”, “light a small”, “kitchen”.
• Everything we learned rebuilding a thirty-year-old kitchen on a weekend budget steps down to 76 pixels and 5lines before it fits — nothing cut, every word smaller: the cost of the sentence.
• The same headline in a box 108pixels tall reaches the example’s 44-pixel floor, keeps 2 lines, ends rebuilding a thirty-year-old k… and is reported as truncated.

The renderer measures real glyphs in its own box with its own floor, so these lines show the fitter’s mechanism, not the tool’s exact output. The reading is the same either way: a smaller size than you set means the words did not fit; an ellipsis means they did not fit even at the renderer’s floor. Neither is a template bug. The fix is fewer words, not a taller box.

Contrast is computed, and one colour is protected

The Studio’s colour doctrine: picked colours are honoured exactly and readability is reported, not enforced. The editor computes the WCAG relative-luminance contrast ratio of each text colour against its background and shows an advisory under 4.5:1, the AA minimum for text (source, 2 October 2026). On the defaults, the overlay title colour #ffffff on the image backdrop #14201c measures 16.8:1 and the accent #3f6f63 on the content background #fbfcfa measures 5.6:1; both pass. A pale yellow such as #f3e9a0 on white measures 1.2:1 and is flagged, not changed.

The thumbnail overlay text is the one exception, because it is drawn over a photograph the design cannot predict. The renderer does not read the photo; it assumes the hardest case for light text, a pure white background, composites its dark scrim over it at 76% opacity to get the protected background #454948, and tests the requested colour against that; a failure becomes the engine’s dark ink #14201c or white, whichever contrasts more. On fixed inputs the default white title is kept at 9.13:1 and a warm yellow #f5d547 at 6.30:1, while the starting accent #3f6f63, requested for the kicker, measures 5.7:1 on bare white yet fails behind the scrim and is drawn in #ffffff at 9.13:1.

Preview at reduced sizes

Guessing how a design degrades is unreliable — checking is cheap. The thumbnail tool builds a 1280×720thumbnail in your browser and shows it in a strip of four reduced widths of its own, labelled home, search, sidebar and mobile, each with a mock duration badge, so you can judge the small version before you upload rather than after the video is live. Those strip widths are the tool’s own preview sizes, not measurements of YouTube; the search and sidebar figures this guide quotes are the site’s wording kept from August 2026, so the two sets need not match, and neither is verified against an official page.

The export offers JPG or PNG, the two formats the help page names. For JPG the tool encodes once at its default quality and, if that file exceeds the tool’s fixed size target, re-encodes once at a lower quality without checking the result again; a PNG export is not size-checked at all. The tool therefore does not guarantee that an export stays under the 2 MBmobile figure — read the file size in the notice the tool shows after the download, and compare it with the spec table above.

The thumbnail as part of a package

A video launch rarely needs only a thumbnail. The same promotion usually wants a 1080×1080 square, a 1080×1350 portrait, and a 1080×1920 story graphic, plus a QR code and captions for each platform. The Studio generates all of that from one destination URL and includes a clean 1280×720 thumbnail in the package ZIP, alongside the QR code as PNG and SVG, a clickable Tap-or-Scan card, a caption set, and a machine-readable manifest. That keeps the thumbnail visually consistent with every other asset in the campaign instead of being designed in isolation.

Those sizes are read from the README the Studio writes into every ZIP. The manifest beside it records, for a fictional campaign: generatedBy, generatedAt, title, destination, routingDestination, campaignType, category, price, shortLink, startsAt, timezone, venue, sourceImage, captionPlatforms, design, designSchemaVersion, files. Its source-image record (contentHash, fileName, resolvedUrl, sourceKind) tells an upload from a pulled YouTube frame, its design record carries the exact colours and sizes used, and design schema version 3 tells the shape from older packages; see package anatomy.

Related guides

• Social image sizes — the current export dimensions for feed, portrait, and story formats across major platforms.
• Promo design basics — hierarchy, contrast, and layout principles that apply to every asset in a campaign, thumbnails included.
• Local image conversion — converting a source photo in the browser before it reaches the canvas.
• On-device background removal — isolating one subject without an upload.
• Methodology — every rule the tools apply, stated from the engines that apply them.

Questions

Is a 1280×720 thumbnail still acceptable to YouTube?

As read on 2 October 2026, the help page sets a minimum of 640 pixels wide (videos); 640 pixels tall for Shorts and asks for 16:9 for videos; the canvas meets both. It now recommends 3840×2160 pixels (16:9) for videos; 2160×3840 for Shorts, so the canvas is a valid size, not the recommended one.

What happened to the 2 MB limit?

The help page no longer states one limit. As read on 2 October 2026: "Limits depend on the device you're using to upload your thumbnail: Mobile: 2 MB for video thumbnails or 10 MB for podcasts. Desktop: 50MB for video, Shorts, and podcast thumbnails." An export under the smaller figure works from either device.

Why did the headline come out smaller than I set, or end in an ellipsis?

The renderer fits the title to the headline box, stepping down from the preferred size until the wrapped lines fit; only at its own minimum size does it keep the lines that fit and add an ellipsis. In the Studio the preferred size is the overlay title you set (default 88 pixels; the control accepts 48 to 120); the thumbnail tool uses the renderer's own per-template preferred size, smaller for Minimal. The floor is the renderer's own constant, below the Studio control's smallest setting. Smaller means the words did not fit; an ellipsis means they did not fit even at that floor. Use fewer words.

Why is my title or kicker white when I picked a different colour?

The overlay text is the one colour the site changes. The renderer assumes a white background, composites its scrim over it at 76% opacity, and tests your colour against that; one under 4.5:1 becomes the engine's dark ink #14201c or white, whichever contrasts more. Every other Studio colour is honoured exactly and only flagged.

Does pulling a YouTube thumbnail or uploading a photo send my image anywhere?

No. The URL becomes a video id in your browser, the platform's own images are fetched directly into the page and checked for size there, an uploaded photo is read locally, and the canvas is rendered, exported and verified on your device.