What the platform does to your upload

Nobody watches the file you uploaded

The earlier lesson established the fact: platforms re-encode everything. This one is about the machinery, because once you know the shape of it you stop guessing about upload settings.

When you upload, the file is decoded and re-encoded into a ladder — a set of renditions at different resolutions and bitrates, so a viewer on a weak connection gets 360p and a viewer on fibre gets 2160p, and the player can switch between them mid-stream without stopping. Your file is the source material for that ladder. It is never itself served to anybody.

What the ladder looks like

A typical ladder runs something like 144p, 240p, 360p, 480p, 720p, 1080p, 1440p, 2160p, each with its own target bitrate. The player picks a rung based on measured bandwidth and screen size, and moves between rungs as conditions change.

Two consequences follow immediately.

First, the rung most people see is not the top one. On mobile data, on a phone screen, a lot of viewing happens at 480p or 720p. Your careful 4K grade is being watched at a quarter of a million pixels. This is an argument for large, clear compositions and readable text, which is a craft decision, not a technical one.

Second, the rungs are not encoded equally. Platforms reserve their better codecs for their more valuable renditions. YouTube encodes popular and higher-resolution content with VP9 and increasingly AV1, both of which look considerably better per bit than H.264, while a low-view 1080p upload may be served as H.264 only. This is the mechanism behind a piece of folklore you will meet constantly.

The 4K upload trick, stated honestly

The advice circulates in this form: "upload your 1080p video as 4K and the 1080p version will look better."

The mechanism is real. Uploading at 2160p puts the file in a class that receives higher bitrate allocations and better codec treatment across the ladder, so the 1080p rendition served from a 4K upload has sometimes measurably outperformed the same video uploaded at 1080p.

Three caveats, and they matter.

It is not documented. Platforms do not publish their allocation rules, which means the behaviour has changed before and will change again.

It costs you an upscale. Rendering 1080p material to a 2160p canvas takes four times the pixels, four times the encode time and four times the upload. The upscale itself adds nothing.

And the size of the gain is small compared to the things you control — exposure, contrast, noise, motion. If your source is clean and generously encoded, you are already collecting most of what is available.

Treat it as a marginal optimisation to test yourself rather than a rule to follow. That is the honest position, and anyone who states it more confidently than that is guessing too.

Processing happens in stages

Immediately after upload, the platform publishes a low-resolution rendition so the video is watchable at once. The higher rungs arrive over the following minutes, and for a long 4K file, over hours.

This is why a video looks terrible the moment you publish it and fine the next morning, and why the first comment on a lot of channels is somebody complaining about quality. Do not re-export, do not re-upload, and do not judge anything in the first half hour. Check the quality menu in the player: if it only offers 360p and 720p, processing has not finished.

Double compression is the real enemy

Every encode throws information away permanently. Encoding twice throws away what the second encoder thinks is unimportant in a picture already stripped of detail by the first, and the artefacts of the first pass — blocking, banding, smeared motion — become structures the second pass faithfully preserves, spending bits on ugliness.

Practical rules that follow:

• Upload from a master, not from a file you already exported for something else.
• Never take a file down from one platform and upload it to another. Export a fresh one.
• Do not pre-compress to speed up an upload. A slow upload once is cheaper than permanent damage.
• If you must send through a messaging app, accept that it will be destroyed and export deliberately small so that at least the damage is done once by you and not twice.

Match frame rate, do not convert it

Upload at the frame rate you edited at. A 25p timeline exported as 30p has had frames duplicated by your editor, and the platform then encodes those duplicates faithfully, producing a permanent stutter that no bitrate fixes.

Platforms handle 24, 25, 30, 50 and 60 correctly. There is no advantage to converting, and one obvious way to ruin the motion.

Vertical and the crop that happens without you

Feeds crop. A 16:9 video in a vertical feed is shown letterboxed or cropped to a square or 4:5 frame depending on the surface, and a 9:16 video shown on a landscape player gets pillarboxed. If a video matters on a particular surface, export for that surface rather than hoping.

Reels, Shorts and TikTok all want 1080 × 1920. Feed posts do better at 4:5 (1080 × 1350) than at 1:1 because it occupies more vertical space on a phone. This is the practical reason to keep an ungraded, uncropped master at full frame: reframing from a master is an export, while reframing from a delivered file is a crop of already-compressed pixels.

Honest limits

Everything specific here — codec choices, resolution tiers, processing behaviour — is observed rather than published, and observed behaviour on a platform is a snapshot. The durable rules are the two that follow from compression itself: encode once, and hand the platform the cleanest thing you can.

Today

Open a video of your own on YouTube, press the quality button and note which rungs exist, then open the statistics-for-nerds panel and read the codec line. If it says vp09, you have seen the ladder choose something better than H.264 for you.