How converting a video to 4K actually works [2026]

CBS spent more than $12 million and over three years turning “Star Trek: The Next Generation” into HD. You’ve got a laptop, a folder of old camcorder files and a weekend. That gap explains most of what goes wrong when people try to upscale video to 4K.

Upscaling software keeps getting better. It still can’t invent detail that was never recorded.

What 4K actually means before you convert anything

When we talk about 4K and 1080p, or ultra-high and high definition, we’re talking about how many pixels PIXEL $0.005171 ▼ -15.5% sit inside each video frame. Ultra-high definition, which is what most people mean when they say 4K, is 3,840 x 2,160 pixels. High definition 1080p is 1,920 x 1,080.

Standard definition is worse than you remember. It’s either 480 or 576 vertical lines depending on your region, and the width shifts depending on how the footage was shot in the first place.

There’s also cinematic 4K, sometimes called “true 4K” or DCI 4K, at 4,096 x 2,160.

The math on the jump from 1080p is the part people skip. For every original pixel that was recorded, you need three more on screen. Four times the pixels, total, and only one in four of them is real.

Why upscaled footage looks soft

You can upscale any output file to 4K. The frame size is just a number. Doing it without losing quality is the hard part, and it’s close to impossible, because changing the resolution doesn’t unlock some hidden sharper version of your file.

Traditional upscaling software uses interpolation to handle the gap, scaling up each frame and working out what it should look like. The whole question is how accurate that guess is. It has to estimate what the missing pixels should be, and it can’t recover real detail that was never captured.

So footage tends to come out softer than the original. The guess-timation used to fill in extra pixels smooths everything over. Sharpening helps around the edges, but it’s a Goldilocks problem: too much or too little and you’ve ruined it. Noise reduction is the same trap, and overdo it and the result looks unnatural.

The Star Trek restoration is the honest benchmark

Professionals converting to 4K start from the best source data they can get, which means the original files. Better source in, better result out. That’s not a preference, it’s the whole ballgame.

The HD restoration of “Star Trek: The Next Generation” shows how far you have to go to do it properly. CBS commissioned the upgrade in 2011 and went back to the show’s original 35mm film negatives, described as having quality equivalent to a 20 megapixel resolution, to remaster all seven seasons from broadcast quality to 1080p for Blu-ray.

The job used 25,000 reels of original film stock, plus visual effects and CGI upgrades along the way. It took more than three years and reportedly cost the studio over $12 million, according to one of its producers.

Here’s the irony. Converting analog film to digital video is the easier version of this job. Film negatives hold far more detail than the original television broadcasts ever carried, so there’s real information waiting to be pulled out.

Start from an older digital master with a fixed resolution and you’re stuck with the pixels you have. When the best source you own is already low resolution, every missing pixel is a guess. Best case, the guess is convincing. It’s still not 100 percent true to form.

Interpolation versus AI, and what each one gets wrong

You probably don’t have $12 million. So the real choice at home is between traditional interpolation and newer AI-based tools. Both create the extra pixels needed to fill a 4K frame. They go about it differently.

Interpolation runs on old-fashioned math. It takes a pixel from a single frame, looks at the color and brightness of its neighbors, then calculates how the surrounding pixels should look. At the crudest level that’s duplication. Better algorithms work to preserve edges and avoid jagged lines.

The output is predictable, which is both the appeal and the limit. Interpolation can’t recreate genuine detail or texture, so the result reads as slightly fake. Your TV may already be doing this on the fly, with results that vary.

AI upscaling attacks the softness problem from a different direction. These models are trained on paired low and high-resolution images, so they learn how common features are supposed to look. Rather than estimating each pixel, the model predicts what a sharper image would be.

Sharper and more convincing, yes. An exact match for the real world, no. And AI upscalers introduce their own strange artifacts that interpolation never produces. The models keep improving, so what looks good today may look dated in a few years.

Old broadcasts bring an extra problem

Working with older video means dealing with interlacing, where frames were split into two halves by odd and even horizontal lines, each refreshing as the screen refreshed. Modern displays don’t work that way. Old broadcasts built for CRT-style screens did.

Those files need de-interlacing before you upscale. Whether a 480i NTSC source can reach 4K and stay watchable is a genuinely open question.

Most upscaling tools, AI or interpolation, give you plenty of settings to push around. If one pass doesn’t land, tweak and run it again. Just make sure you’ve got the disk space, because you’ll be doing this more than once.

Resolution isn’t the only thing holding your file back

Don’t drop a 1080p file into an upscaler and expect a clean 4K result. You’ll get 4K in the technical sense, since the frame is bigger. Getting there without losing quality takes several attempts and a tolerance for disappointment.

Short leaps work better than long ones. Going from 1440p to 4K needs far fewer fake pixels than 1080p to 4K, and vastly fewer than 480p to 4K. The software is guessing at the missing pieces either way. AI makes the guessing easier and potentially better, but a like-for-like match isn’t guaranteed.

Then there’s bitrate, which nobody thinks about until it bites. Files get compressed to save disk space, meaning less data is stored for every second of video. Upscaling to 4K doesn’t undo that. Blocking and banding from a low bitrate source stay right where they are, only larger.

Dig out the best source you own before you install anything. If that file is good enough, the rest of this gets a lot easier.