How to avoid macroblocking in dark scenes.
Eight steps before you upload, and one structural fix.

The first coefficient of a block, the DC term, is its average brightness. Everything after it describes how the block varies inside itself: the gradient, the edge, the texture. When the step size gets large enough, every one of those later coefficients rounds to zero and only the average survives. The decoder then has nothing to draw but a rectangle of one flat value. Put a few hundred of those side by side across a falling gradient and you get the staircase of squares people call macroblocking, with edges that line up with the codec’s grid rather than with anything in the photograph.

Dark areas lose that fight first, for two reasons that compound. The first is perceptual. Rate control is steered by a model of what a viewer notices, and that model says shadow detail is cheap: a viewer will forgive a soft corner of a night frame long before they forgive a smeared face. So the encoder is instructed, quite deliberately, to raise the step size where the picture is dark. The second reason is arithmetic. There is genuinely less signal down there. A slow falloff from near-black to slightly-less-black produces small coefficients to begin with, so it takes very little coarsening to round the whole set to zero. The bright half of your frame has to be starved hard before it visibly breaks. The bottom two stops break at the first sign of pressure.

This is also why the problem shows up in motion and not on the frame you exported as a still. Between the full frames, the codec is coding differences, and in a dark scene those differences are tiny. Coarse quantisation zeroes them, the block holds its old value for a few frames, then updates all at once. What you see is not a static pattern but a crawl.

Naming it correctly saves a wasted evening. If the artifact follows a grid, work on the encode. If it follows the gradient, work on bit depth and dither, which is a different guide: see what happens to film grain after upload for the noise side of the same problem. If it swarms around your lower third, your titles are being encoded at the same starved bitrate as everything else.

Vimeo states it plainly: during transcoding, your file “is re-encoded into several formats that will be available to view on Vimeo”. It then publishes the maximum video bitrate for each of those formats, which is unusually candid and extremely useful. On the H.264 ladder the ceilings are 0.3 Mbps at 240p, 0.5 at 360p, 1.5 at 540p, 2.5 at 720p, 7 at 1080p60, 12 at 2K and 22 Mbps at 4K. On the HDR and above-4K HEVC ladder they are 2.2 at 720p, 5 at 1080p, 10 at 2K and 16 Mbps at 4K. Vimeo also attaches the qualifier that matters most here: those figures are a maximum, and “your video could transcode at a lower bitrate depending on the complexity of the image”.

A night exterior with grain is, by any encoder’s definition, a complex image. So the published ceiling is the best case for the scene least likely to reach it.

YouTube is the opposite kind of problem. It says outright that it “always re-encodes videos to optimize their playback quality”, and that processing starts in low quality with the higher resolutions finishing later, which is why a 4K upload can look coarse for hours after it goes live. What YouTube does not publish, anywhere, is the bitrate it delivers at any resolution. You cannot plan against a number that does not exist.

For scale outside this argument, Apple’s own HLS authoring specification, the document the streaming industry builds ladders against, tops out its 4K HEVC example rung at 16,800 kbps SDR. Streaming as a discipline is built around a ceiling in that region. That is not a failure of Vimeo or YouTube. It is what adaptive delivery is for, and it is why a master that carries its shadow gradation comfortably at 150 Mbps has to give most of it up to fit through the door.

Most people check the wrong thing in the wrong conditions and conclude the problem is fixed. Three rules make the test worth doing.

Turn the room lights down. Shadow artifacts are invisible in a bright room and obvious in a dark one, and your client is watching at 11pm with the lights off. Then view at 100 percent, one image pixel to one screen pixel. A 4K file scaled into a half-width browser window is being resampled by the browser, and resampling smooths the block edges you are looking for. Full screen on a 4K display, or a 1:1 window, or you are not measuring anything.

Then, and this is the one people skip, watch a moving shadow rather than pausing on a still. Macroblocking is at its most visible when a block holds a stale value for several frames and then snaps, which a paused frame cannot show you. Find a shot where the camera drifts across a dark wall, or where a figure moves through an unlit part of the frame, and play it at speed. If the shadow appears to be built out of tiles that update on different schedules, you still have work to do.

Last: judge on the rendition your client will actually receive. If you are delivering through an adaptive platform, step the player’s quality selector down to 1080p and look there, because a laptop on hotel wifi is not getting the top rung. Signing off on the 4K rendition and shipping the link is how a scene that passes your review fails theirs. If you are delivering a single file that is not re-encoded, the rendition question does not arise and what you approved is what plays.

uncompressed.io streams the exact bytes you upload. There is no transcode anywhere in the path, no ladder, and no re-encode of video or audio, so the shadow gradation the viewer sees is the gradation you exported, block for block. If your night exterior is clean at 150 Mbps in the suite, that is the file the client opens. Uploads are MP4 or MOV in H.264, HEVC or AV1, up to 50 GB per file. See video bitrate explained for how to pick the number in the first place.

Two caveats belong on the same page as that claim, because a page that only lists the upside is an advertisement. The first is playback: whether a given file decodes is the viewer’s hardware decoder’s business, and browser vendors document what they accelerate in fairly narrow terms. High-bitrate and high-bit-depth files sit closer to the edge of what consumer hardware handles comfortably, and 10-bit 4:2:2 decode in a browser is not documented by any browser vendor, in either direction. We store and serve the file untouched. We cannot promise you what a stranger’s laptop will do with it.

The second is the missing ladder. Because there is only one rendition, a viewer on a slow connection buffers instead of silently dropping to a softer version. That is the honest trade, and it genuinely cuts both ways: nobody ever sees a mushy rendition of your work, and somebody on bad wifi may see a spinner. If your priority is that every viewer gets something playable immediately, an adaptive platform is doing a job this one deliberately does not do.

Say your delivery target is a public launch film, or a campaign that will be watched on phones on mobile data across a whole country. You want the adaptive ladder. You want the 240p and 540p rungs, and those rungs are low bitrate on purpose, because their job is to play at all rather than to play beautifully. Vimeo publishes 0.3 Mbps at 240p and 1.5 Mbps at 540p on its H.264 ladder. No export setting on earth keeps a grainy night exterior clean at 0.3 Mbps. It will block, and that is the system working as designed.

The fix in that case is not a different host. It is the grade. Lift the floor further than you would for a cinema deliverable, keep the shadows simpler, put less texture and less movement in the parts of the frame that are going to be starved, and light so that the story of the shot survives at low bitrate. Colourists who cut for broadcast and social have done this for years, and they do it in the suite, not in the export dialog.

The same honesty applies to platform choice. If the goal is reach and discovery, YouTube is free, has an audience nobody else has, and no private host competes with it on that. If the goal is a public marketing page with a polished player, marketing tooling and OTT products, Vimeo has an entire product suite that this site has no equivalent to. Where this site is the better answer is narrower and specific: a graded film, going to a named audience, that has to look on their screen exactly the way it looked on yours.