A thumbnail has to survive being shrunk to the size of a postage stamp on a phone, sitting next to eleven other thumbnails, viewed by someone who is not paying attention. Almost every thumbnail mistake comes from designing it at full size on a large monitor and never checking what happens at 320 pixels wide.
Here are the numbers, the formats, and a research workflow.
The specifications
Upload your thumbnail at 1280 × 720. That is 16:9, and it is what YouTube stores as the maximum-resolution version.
- Minimum width: 640 pixels. Below that YouTube may reject it, and it will look soft everywhere.
- Aspect ratio: 16:9. Anything else gets letterboxed or cropped.
- File size limit: 2 MB.
- Formats: JPG, PNG, GIF, BMP. Use JPG for photographic thumbnails, PNG when you have flat colour and text and want crisp edges.
Going above 1280 × 720 gains you nothing — it is downscaled. Going below costs you sharpness on TV and desktop, where thumbnails render large.
The sizes YouTube actually serves
Your one upload gets resized into a family of images, each used in a different place:
| Name | Dimensions | Where it appears |
|---|---|---|
maxresdefault | 1280 × 720 | TV, large desktop players, embeds |
sddefault | 640 × 480 | Some legacy surfaces |
hqdefault | 480 × 360 | Search results, sidebars |
mqdefault | 320 × 180 | Mobile feed cards |
default | 120 × 90 | Small lists, playlists |
The important line is mqdefault. 320 × 180 is where most people see your thumbnail, and it is why the design rules below exist.
Before uploading, shrink your thumbnail to 320 pixels wide and look at it. If you cannot tell what it is in one second at that size, the design has failed regardless of how it looks at full resolution.
Design rules that follow from 320 pixels
Three or four words maximum. Not a sentence. The text is a hook, not a description — the title is right underneath doing that job.
Text at least 90 pixels tall in the 1280 × 720 canvas. That is roughly 22 pixels once shrunk to mobile: readable. Anything under 60 pixels disappears.
One subject. A face, an object, a before/after pair. Three things competing for attention at postage-stamp size resolve to zero things.
High contrast, and check it in greyscale. Convert a copy to greyscale — if the subject stops separating from the background, the colour was doing all the work and it will fail against a dark or light UI.
Keep the bottom-right corner clear. The duration badge sits there and covers whatever you put underneath it.
Do not repeat the title. Title and thumbnail are one unit with two slots; spending both on the same fact halves your surface area. More on splitting that work in the titles guide.
A thumbnail that promises something the video does not deliver raises click-through rate and destroys retention — and retention is what decides whether YouTube keeps showing the video after the first few hours. See how ranking actually works for why that trade is a bad one.
Grabbing a thumbnail from any video
Every public video exposes its thumbnails at predictable URLs. Take the video ID — the part after v= in the watch URL — and:
https://img.youtube.com/vi/VIDEO_ID/maxresdefault.jpg
https://img.youtube.com/vi/VIDEO_ID/hqdefault.jpg
https://img.youtube.com/vi/VIDEO_ID/mqdefault.jpg
https://img.youtube.com/vi/VIDEO_ID/default.jpgTwo things to know. First, maxresdefault.jpg does not always exist — it is only generated when the source video was uploaded at 720p or higher, so older or low-resolution videos will 404 there and you have to fall back to hqdefault. Second, i.ytimg.com serves the same images and is what YouTube's own pages use.
If you would rather not construct URLs by hand or guess which resolutions exist, paste the video link into the thumbnail downloader — it resolves the ID, checks which sizes are actually available, and gives you each one. The same applies to channel branding: profile pictures and channel banners have their own resolution families.
Downloading a thumbnail is fine for research, competitive analysis, and reference. Republishing someone else's thumbnail as your own is copyright infringement. Study the composition, not the pixels.
Running a competitor thumbnail audit
This is a 20-minute exercise that changes how you design, and most creators have never done it.
- Search your target query and take the top ten results.
- Download all ten thumbnails and drop them into one grid — a slide, a Figma frame, anything.
- Shrink the whole grid until each one is about 320 pixels wide. This simulates the actual feed.
- Look for the pattern. Almost always, eight of the ten are the same: same colour palette, same face-plus-text layout, same arrow.
- Find the gap. If everyone uses a red-and-yellow palette, a clean dark thumbnail stands out. If everyone shows a face, an object shot does. Standing out is a relative property — it only exists against the specific grid your video will appear in.
- Read the pairs. Note which titles and thumbnails split the work versus repeating each other. The ones that split are usually the ones with the best positions.
Repeat this per topic, not per channel. The competitive grid for "best budget microphone" looks nothing like the one for "microphone not working in OBS", even on the same channel.
Testing after publishing
Thumbnails are replaceable at any time, and swapping one does not reset a video's performance.
A sane cycle:
- Publish, wait 7–14 days for the CTR figure to stabilise.
- Compare against your channel's median CTR for the same traffic source — Browse and Search have very different baselines, and comparing across them is meaningless.
- If it is below median, change the thumbnail. Only the thumbnail.
- Wait another 14 days and compare the same window.
Change the title at the same time and you learn nothing about either. If your channel has A/B thumbnail testing available in Studio, use it — it is a much cleaner experiment than sequential swaps, which are contaminated by the video's age.
The short version
- Upload at 1280 × 720, 16:9, under 2 MB, JPG or PNG.
- Design for 320 pixels wide, because that is where most people see it.
- Three or four words, one subject, high contrast, check it in greyscale.
- Keep the bottom-right corner clear for the duration badge.
maxresdefault.jpgmay not exist; fall back tohqdefault.jpg.- Audit the competitive grid per query, not per channel.
- Test one variable at a time, over 14-day windows.


