YouTube Video Thumbnail Download: Which File You Actually Get

The file name in the address is a request, not a promise. Measured on 9 October 2026 across five videos and five names each: maxresdefault.jpg and hq720.jpg came back as exactly the same bytes — same hash, same size, same pixels — in three of the five videos. On a fourth video the opposite happened: hq720.jpg returned 179,638 bytes while maxresdefault.jpg returned 108,323, so the name with "max" in it was 71,315 bytes smaller. Downloading the "best" name therefore proves nothing on its own; the byte count does. Here is the full table, the two names that collide, and the check to run after a download.

What each name returned

Five videos, five file names, 25 requests. Bytes and pixel dimensions are the downloaded file's own:

Name requested	Video A	Video B	Video C	Video D	Video E
`maxresdefault.jpg`	65,324 1280 × 720	108,323 1278 × 720	184,051 1280 × 720	404 1,097	82,327 1280 × 720
`hq720.jpg`	65,324 1280 × 720	179,638 1280 × 720	184,051 1280 × 720	404 1,097	82,327 1280 × 720
`sddefault.jpg`	31,029 640 × 480	46,793 640 × 480	84,176 640 × 480	404 1,097	40,116 640 × 480
`hqdefault.jpg`	21,011 480 × 360	32,195 480 × 360	53,433 480 × 360	15,921 480 × 360	25,889 480 × 360
`default.jpg`	2,888 120 × 90	4,120 120 × 90	5,096 120 × 90	4,523 120 × 90	3,963 120 × 90

The bold cells are the ones worth reading twice: on videos A, C and E the two "high quality" names produced identical byte counts — 65,324, 184,051 and 82,327 — and the hashes matched too, so these were not coincidences but the same file served under two names. On video B they diverged, and it was the lower name that was larger.

Flat vector diagram: two file cards with different name labels side by side, both joined by a bracket to a single identical picture card, while a third smaller card on the right is linked to a visibly larger picture card, suggesting that different names can hand back the same file and that the larger file is not always the one with the bigger name — Two names, one file on the left. On the right, the case that breaks the rule: the smaller-sounding name returned the larger file.

The two names that collide

Across the five videos measured, the pair involved in every collision was maxresdefault.jpg and hq720.jpg. When a video has no separate 1280 × 720 rendition beyond the one already used for maxresdefault, the host serves the same bytes for both names. You can see it without any tools: both downloads show the same size on disk, and the same two numbers in the dimensions.

The practical reading: if your workflow downloads maxresdefault.jpg and falls back to hq720.jpg, the fallback may be pointless on that video — and if you download both "to be safe", you may be storing the same file twice. Compare the two files' sizes (or hashes) after downloading; if they match, keep one.

The case where the "max" name is the smaller file

On video B in the table, maxresdefault.jpg returned 108,323 bytes at 1278 × 720 — note the odd 1278 width — while hq720.jpg returned 179,638 bytes at 1280 × 720. Same video, same moment, 71,315 bytes apart, in favour of the name that sounds like the fallback.

That is the clearest argument for judging a download by its numbers rather than its name. A 1278-pixel-wide rendition looks like a re-encode, and here it also came back smaller than the 1280-wide one. The file you want is the one with more bytes, not the one with the more impressive name.

Where the grey square fits in

One column of the table returned the placeholder rather than a picture: on video D, maxresdefault, hq720 and sddefault all answered 404 with a 1,097-byte, 120 × 90 file. That is the same placeholder described on the verification page — a real, valid JPEG that is not the thumbnail. If your download "succeeded" but the file is about a kilobyte, you saved that, not the picture.

A download routine that does not trust names

Ask for the two candidates: maxresdefault.jpg and hq720.jpg. One of them is often the whole answer, and you cannot tell which from the name.
Compare the byte counts. If they match, you have the same file twice — keep one. If they differ, keep the larger.
Discard anything near 1 KB. That is the placeholder, and it arrives with a 404 that a browser may not surface.
Fall back in size order — sddefault.jpg, then hqdefault.jpg — only when both candidates came back as the placeholder. The availability page shows how often that happens.
Read the dimensions you got, not the ones you expected: 1278 × 720 turned up in this run where 1280 × 720 was the assumption.

What you may then do with the file

Downloading is the easy half; the picture still belongs to the uploader. The platform's own terms, quoted with their section headings, are on the usage page. Nothing here stores, mirrors or re-uploads a thumbnail, and this is not legal advice.

Questions people actually ask

"I downloaded the highest quality name — did I get the best file?"
Not necessarily. In this run hq720.jpg was 71,315 bytes larger than maxresdefault.jpg on one video, and identical on three others. Compare the byte counts and keep the larger file.

"Why are my two downloads the same size?"
Because the host served the same file for both names. maxresdefault.jpg and hq720.jpg matched byte-for-byte in three of the five videos measured — same bytes, same 1280 × 720 pixels. Keep one copy.

"Does the download tell me when it failed?"
No. A missing name returns a real 1,097-byte JPEG, so the download completes successfully. Check the size: about a kilobyte means you saved the placeholder, not the thumbnail.

"Can the same video give me a 1278-pixel-wide file?"
Yes — one measured rendition was 1278 × 720 while others for the same video were 1280 × 720. Read the dimensions of the file you received rather than assuming the common 1280 × 720.

"So which single name should I ask for first?"
Either of the two high-quality names, then compare. If you only want one rule: download both, keep the one with more bytes, and treat anything near 1 KB as the miss.

All numbers on this page were measured on 9 October 2026 by requesting 25 URLs directly from img.youtube.com — five public videos, five file names each — recording the HTTP status, byte count, Content-Type and SHA-1 of every response body, and reading the pixel dimensions out of each returned file's own JPEG header. Every successful response was served as image/jpeg. Videos are identified as A–E only; no titles, channels or watch links appear on this page, and none of the images are hosted here. Another sample of videos would change the counts, which is why the names of the colliding pair matter more than the totals.