The pre-upload thumbnail checklist
Every item here is a threshold somebody measured, not a preference: the size your design is delivered at, the character count a title survives to, the share of the frame a face needs. Run them in order and the expensive mistakes are gone before the upload button.
Why a list of numbers and not of advice
Thumbnail advice is mostly unfalsifiable — use a face, add contrast, stand out — and unfalsifiable advice cannot be checked before publishing, which is exactly when you need it. Each line below is a number instead, and each number came from measuring the interface your design will land in.
The list is short because most of the guidance in this field collapses into a handful of constraints once you measure it. Everything here links to the guide that derived it, so nothing has to be taken on trust. YouTube publishes its own, much shorter, thumbnail and title tips; this list is what is left once you insist on numbers.
The floors, in one table
Nine measurements and where they come from. They are floors, not targets: clearing them does not make a design good, but failing one makes it fail in a way nobody can see in an editor.
| What | The number | Measured in |
|---|---|---|
| Master file | 1280 × 720 or larger, 16:9, JPG or PNG | size and specs |
| What is actually delivered | 336 × 188 in a suggested column on a 2× screen | why it looks blurry |
| Title that survives | 39 characters in the desktop suggested column | thumbnail and title |
| Face size | a third of frame height on a card, over half in suggested | how big a face has to be |
| Text size | smallest text at least a tenth of image height | how big text has to be |
| Keep clear | bottom-right corner and the bottom edge | the safe zone |
| Separation | brightness, not hue — the corpus median is only 1.57:1 | colour blindness |
| Edge in dark mode | a light edge or a light subject against #0F0F0F | dark mode |
| Your row | the niche median it will sit against | standing out |
Two of those nine are not about your design at all — the delivered size and the row — and they are the two that decide whether the other seven mattered.
The order to run them in
- Export once, at 1280 × 720From the lossless working file straight to the upload, in a single pass, without a second trip through a compressor. Every intermediate JPEG is a round of artefacts the platform's own encoder will amplify rather than clean up.
- Look at it at a tenth of the sizeScale the design to about 15% and read it. Whatever you cannot make out there is not in the thumbnail as far as a viewer is concerned, because that is roughly the size a suggested item is delivered at.
- Desaturate itOne slider. If the parts stop separating in greyscale, the design is leaning on hue — which disappears for red-green colour blindness and thins out at delivery size for everyone else.
- Cover the bottom-right corner and the bottom edgeTwo fingers, or the overlay template from the safe zone guide. The duration badge and the watched-progress bar are drawn there on almost every surface.
- Put it on a dark backgroundMost viewing happens in dark mode, and a dark frame against #0F0F0F loses its boundary. If the edges dissolve, the design needs a light edge or a lighter subject, not more saturation.
- Put it in the row it will appear inLoad your niche's pack, or import your competitors as a CSV, and look at the grid. This is the only check that answers the question the others cannot: is yours the cell that reads differently?
- Read your own title at 39 charactersIn the suggested column the title is cut to about 39 characters. If the point of yours arrives after that, the pair is carrying less than you think — and the thumbnail has to carry more.
Why the order is the order
The first three are cheap and catch the most: an export mistake, a design that only works large, and a separation built on hue. None of them needs anything but the file you already have.
The last two are the ones people skip, and they are the ones that involve other people's work — the interface drawing over your frame, and fifty other thumbnails competing with it. That is why they come last: they are the only checks that cannot be done on the design alone.
The two surfaces that govern
Almost every number above is set by the same two places: the suggested column beside a playing video, on desktop and on a phone. They are the smallest surfaces a viewer will see you on — 168 × 94 and 143 × 80 — and the ones where a decision is made about what to watch next.
Everything else is more generous. A phone feed card is 293 × 165 and takes 14.7% of the screen, which is measured in the guide on mobile and is the largest a thumbnail ever gets. Design for the two narrow columns and the wide surfaces take care of themselves.
Three things this list deliberately leaves out
Click-through rate. Nothing you can check before publishing predicts it, and a number from someone else's channel is not a benchmark. What CTR does and does not measure is in its own guide.
Emotion. Whether an expression is compelling is not something a measurement of the interface can answer. What it can answer is whether the expression survives the reduction, which is a size question and is on the list.
Style. No item here says what a thumbnail should look like. They say what it must survive. The space left over is the actual work, and it is the part a checklist cannot do for you.
Four rules that fail their own measurement
“Use red to stand out.” Red is the leading hue in eight of eleven niches, up to 57% of the saturated pixel weight. In most rows it is camouflage — measured in the niche guide.
“Fill the file size limit for quality.” The limit — 50 MB from a computer since it was raised from 2 MB, and 2 MB only from the phone — is a ceiling on the upload, not a delivery budget: what viewers receive is between 7 KB and 318 KB. Compressing twice to reach it is the only part that changes anything, and it changes it for the worse.
“More contrast makes you stand out.” Internal luminance contrast is nearly identical across all eleven niches, between 0.228 and 0.269. It is a requirement, not a differentiator.
“Keep every thumbnail identical.” Across 33 measured channels, brightness varies inside a channel almost as much as across its whole niche. What they hold constant is palette and subject placement — the consistency guide has the split.
The one adjustment worth making per niche
The floors are the same everywhere; the row is not. Median brightness across the eleven niches runs from 0.178 in Science to 0.375 in Fashion & Beauty, so “bright enough” means opposite things at the two ends.
The cheapest way to make that concrete is to look at it: open the tool on the gaming pack, or swap the name for any other — ?pack=finance, ?pack=foods — and drop your design in. The gaming guide is the same measurement run on one row, if you want the shape of the answer before you look.
After the upload, one number matters
The change is stored immediately, and then nothing appears to happen: YouTube's image CDN sends cache-control: max-age=7200 on an URL with no version in it, so every cache may keep the old image for two hours. Measured in the guide on that, along with the way to see the stored file right now.
Which is the argument for the whole list. The window in which a thumbnail decision is cheap closes at upload, and reopens two hours later with the first impressions already spent.
The one-minute version
Export once at 1280 × 720. Shrink it to 15% and see what is left. Desaturate it and see what still separates. Keep the bottom-right corner empty. Then put it in the row it will compete in — that last one is what the preview tool exists for, and it catches what the other four cannot. Seeing it where it will compete, rather than in the editor, is a step of its own: how to preview it before publishing covers the four ways and what each one can tell you.
The questions everyone asks
What should I check before uploading a thumbnail?
Nine measured floors: 1280 × 720 or better as JPG or PNG, legible at a tenth of the size, separation by brightness rather than hue, the bottom-right corner clear, an edge that survives dark mode, a face at a third of the frame, text at a tenth of the height, a title whose point lands in 39 characters, and the row it will appear in.
What is the single most useful check?
Looking at the design at the size it is delivered — about 336 pixels wide on a suggested item — because it catches text, faces and contrast in one pass. The second is looking at it beside the competitors it will actually sit next to.
Do these rules change by niche?
The floors do not; the row does. Median brightness across eleven niches runs from 0.178 to 0.375, so what counts as unusual is different at the two ends. Load your niche's pack and the difference is visible rather than theoretical.
How long does the whole check take?
Under five minutes for the seven steps, and about one for the short version. Both are cheaper than the two-hour cache window a post-upload thumbnail swap costs you.
Is there anything a checklist cannot tell me?
Whether the idea is any good. Every item here is about survival — of a reduction, of an interface, of a row of rivals. What to put in the space that survives is the part no measurement decides.