TV Upscaling Explained: What Changes and What Does Not

What does TV upscaling actually do?

TV upscaling is the process of stretching a lower-resolution picture across a screen with more pixels, and estimating values for the pixels the source never supplied. A 4K television has 3,840 by 2,160 pixel positions; a 1080p signal brings 1,920 by 1,080. The scaler fills the gap by calculation. Done well, edges look clean and the picture fills the screen without blockiness. What it cannot do is recover fine detail that was never recorded, or that compression threw away before the signal arrived.

That is the whole idea in one paragraph. The rest of this page is about the parts that make it interesting: where in your setup the scaling happens, which other processing rides along with it, what the newer "AI" scalers are really doing, and how to tell whether one path looks better than another without fooling yourself.

Why does a TV have to upscale at all?

A modern flat panel is a fixed grid. Every pixel position on a 4K screen has to be told what to show on every frame, whether the program supplied that information or not. A smaller picture shown pixel for pixel would sit in a sharp little rectangle in the middle of the glass, so anything below the panel's native resolution is normally resized to fill it.

The arithmetic shows how much work that is. A 1080p frame holds 2,073,600 pixels; a 4K UHD frame holds 8,294,400, exactly four times as many. RED Digital Cinema describes the jump from 1080p to 4K as four times the image information, which matches that count. Older material asks for far more:

On an 8K panel (7,680 by 4,320, or 33,177,600 pixels), a 1080p source supplies 1 in 16 and a 4K source 1 in 4. The bigger the gap, the more of the final picture is the scaler's estimate rather than the camera's record.

Where does scaling happen in the chain?

Scaling happens once, somewhere between the source and the panel, and it is not always the television doing it. NVIDIA's explainer notes that upscaling is carried out by the playback device, such as a smart TV or a streaming media player. In practice the candidates are:

• The TV itself, when it receives a signal below its native resolution, including from its own built-in apps.
• A streaming box, disc player or games console set to output the panel's resolution regardless of what it is playing.
• An AV receiver sitting between source and TV, if its video processing is switched on.

This matters because the TV's information screen reports what arrived at its input, not what the program originally was. A box set to output 2160p will hand the television a 4K signal built from a 1080p stream, and the TV has nothing left to scale. If you have never checked which device does the job in your setup, the output-resolution menu on each source is the place to look. Our 4K vs 1080p guide walks through that specific case in more detail.

Which other processing runs alongside the scaler?

"Upscaling" on a spec sheet usually means a bundle of steps, and several of them change the picture more visibly than the resizing does.

Deinterlacing. Some broadcast and older disc formats are interlaced, meaning each frame is sent as two half-pictures of alternating lines. Deinterlacing is the process of converting interlaced video into a non-interlaced or progressive form, and it is built into most modern disc players and televisions. Get it wrong and moving objects show combing, a comb-tooth fringe along their edges.

Noise reduction. Many sets have separate controls for random grain and for compression noise. One Sony BRAVIA help guide, for instance, lists an MPEG Noise Reduction setting described as reducing noise in MPEG-compressed videos, next to a general Noise Reduction control and a Reality Creation setting that "Adjusts detail and noise for a realistic picture."

Sharpening and detail enhancement. These boost contrast along edges so the image reads as crisper. A little can offset the softening that interpolation introduces; a lot draws bright outlines around everything.

Each of these is a separate decision with its own trade-off, which is why two televisions with the same panel resolution can make the same 1080p broadcast look noticeably different.

How does interpolation fill in the missing pixels?

Interpolation means estimating unknown values from known neighbours. RED's guide puts it simply: upscaling uses known pixel values to estimate unknown intermediate pixels. The methods differ in how many neighbours they consult and how they weigh them:

• Nearest-neighbour copies the closest source pixel. It is fast and keeps hard edges, but the image scaling literature notes it can introduce jaggedness in previously smooth images.
• Bilinear blends the four nearest pixels. It is smoother, but the same reference says it reduces contrast at sharp edges, which softens fine lines and text.
• Bicubic and Lanczos look at a wider neighbourhood. They generally hold edges better, at a higher computing cost.
• Edge-directed methods first work out which way an edge runs, then interpolate along it rather than across it, to avoid staircase steps on diagonals.

None of these is free. RED states that all methods incur some combination of blurring, blocking and halo artifacts, and that reducing one tends to increase another. That is the honest shape of the job: a scaler is choosing which flaw to make least visible, not removing flaws altogether.

What is AI upscaling, and how is it different?

AI upscaling replaces a fixed formula with a model trained on example images. NVIDIA describes it this way: given a low-resolution image, a deep learning model predicts a high-resolution image that would downscale to look like the original, and the network has to be trained on large numbers of images to do that accurately. Instead of averaging neighbours, the model draws on what textures and edges usually look like.

That can produce cleaner edges and more convincing texture than a classic scaler, particularly on well-encoded HD. It also has a catch built into its design. The image scaling reference notes that AI tools inevitably have to invent, or at least recreate, details that were or were not there. Fine texture that looks plausible is not the same as texture that was recorded. On a nature documentary that may not bother you; on a face, a sign or a printed page, invented detail can look subtly wrong.

AI processing also has limits on what it accepts. NVIDIA's support page for its SHIELD streaming player, for example, says its AI mode is not available for video below 480p, above 30Hz, or in an RGB color space, and that such content falls back to its conventional "Enhanced" mode. Other devices set their own boundaries, so check the manual for yours rather than assuming the AI label applies to everything you watch.

What can upscaling never bring back?

Information that was lost before the signal reached the scaler stays lost. Three common causes:

1. Capture limits. A program shot or mastered at a lower resolution simply holds less fine detail. RED's comparison notes that upscaled footage still falls short when it comes to depicting more detail than native 4K, with fabric, hair and landscape texture the usual giveaways.
2. Compression. Streaming and broadcast video is compressed with lossy methods that discard some of the media's data to fit the available bandwidth. What was discarded is not in the signal, and the scaler can only smooth over the damage. The same source notes that deblocking filters can reduce artifacts at block boundaries, but that no post-processing technique has been shown to improve quality in all cases.
3. Earlier processing. A picture that was already sharpened, noise-reduced or scaled once upstream carries those decisions with it. A second scaler cannot undo them.

The practical consequence, in our editorial judgement, is that a cleanly encoded source gives any scaler more to work with than a heavily compressed one at the same resolution. A soft, blocky stream can be smoothed on its way to the panel, but not restored. Resolution on the label is only one input.

Does upscaling improve picture quality?

Compared with showing a small picture unscaled, yes: it fills the screen and hides the pixel grid, and NVIDIA notes that the plainest copy-and-repeat method can often appear muted or blurry, which is the gap better scalers try to close. Compared with native material at the panel's resolution, no. RED's conclusion is that upscaled 4K still does not compare to true 4K resolution.

The useful question sits between those two answers: does this scaler, on this source, from your seat, make the picture look more like the original or less? Heavy sharpening can make a picture look "more detailed" on first glance while adding halos and edge shimmer that grow tiresome over an evening. Strong noise reduction can make a grainy film look clean while smearing skin into wax. Better is the version that stays truthful to the source, not the one that shouts loudest.

How do you compare two upscaling paths fairly?

Whether you are weighing a streaming box against the TV's own scaler, or an AI mode against a standard one, the comparison only means something if one thing changes at a time.

1. Pick one short, familiar passage with faces, fine texture, small text and some motion. Use the same title from the same service or disc.
2. Fix everything else. Same HDMI port, same picture mode, same brightness and contrast settings, same seat, same room lighting.
3. Change only the scaling path. For example, set the source to output 1080p so the TV scales, then set it to 2160p so the source scales, and replay the identical passage.
4. Use built-in comparisons where they exist. NVIDIA's SHIELD offers a side-by-side demo mode comparing Basic and AI upscaling in real time; similar split-screen tools on other devices make the same test easier.
5. Look for specific faults, not a general impression. Check edges for bright halos, diagonals for stair steps, skin for a waxy smear, dark scenes for blocky noise, and subtitles for ringing.
6. Watch for a few minutes, then switch back. A first-glance preference for the sharper image often reverses once halos and shimmer start drawing the eye.

If you are judging televisions in a shop instead of at home, the same principle applies with less control; our guide to comparing TVs in a store fairly covers what you can and cannot read from a showroom wall.

Which upscaling-related settings are worth adjusting?

Start from an accurate picture mode, then treat detail and noise controls as seasoning. Our first picture settings guide covers that starting point. A few habits help:

• Sharpness is easiest to judge on text and fine lines; raise it only until edges look clean, and back off at the first sign of outlining.
• Noise reduction is most useful on poor sources such as old recordings or low-bitrate streams, and usually unnecessary on a clean disc.
• AI or "super resolution" style modes are worth testing both on and off with the method above, since their effect varies by source.
• Game modes often trim processing to cut delay, so gaming pictures may look less polished than film modes; our game mode and input lag guide explains that trade.

Upscaling is not magic and it is not a con. It is a set of estimates, made at speed, about pixels that were never there. Once you know that, the right settings are simply the ones that keep those estimates quiet.

Sources

• RED Digital Cinema, "Upscaled 1080P vs 4K"
• NVIDIA Blog, "What Is AI Upscaling?"
• NVIDIA SHIELD Support, "AI Upscaling"
• Sony BRAVIA Help Guide, "Adjusting various picture settings"
• Wikipedia, "Image scaling"
• Wikipedia, "Deinterlacing"
• Wikipedia, "Compression artifact"