Upscaling Video to 4K: An Honest Guide

A 4K television is on the wall. A 1080p video is on the screen. It looks soft, and you wonder whether turning the file into 4K would fix it. Or you are a YouTuber who heard that uploading in 4K gives better quality even if you filmed in 1080p. Or a client asked for “4K deliverables” and all you have is a 720p screen capture.

These are three different situations with three different answers, and the honest answer for each is “it depends on what you want from it”. This article goes through them, with real numbers from tests I ran on the box, and ends with the settings I would use.

What upscaling to 4K actually does

A 4K frame (3840 by 2160) has 8.3 million pixels. A 1080p frame has 2.1 million. So going from 1080p to 4K means that out of every four pixels, three are computed from the one you had. That works out the same whichever filter you use. The filter only decides how.

The information content of the image does not change. If your 1080p video had a certain amount of detail, the 4K version has the same detail spread over more pixels. A good filter keeps edges clean and avoids jagged lines. That is real, and it is the whole benefit.

There is one more benefit that has nothing to do with sharpness: your TV or monitor already does this when you play 1080p on a 4K display. It scales the picture itself, using whatever filter its chip includes. The filter in the TV is usually decent and sometimes excellent, since TV makers spend a lot on it. So the question “should I upscale it myself” often becomes “is my filter better than my TV’s”.

Numbers from my test

In the test I described in the low quality to HD article, I shrank a 1080p clip and scaled it back using several filters. The same ordering holds when going up from 1080p to 4K, and the gap between filters is smaller in that case because you start from more information. From a 480p start, bicubic reached 33.4 dB PSNR and Lanczos 33.6 dB. From 360p, Lanczos reached 31.8 dB. In both cases, nearest neighbor sat about two dB lower.

What that means for 4K: any decent filter is fine. Choose Lanczos or bicubic, avoid nearest neighbor and plain bilinear, and do not expect a difference you can see across the room.

Three situations, three answers

1. You want it to look better on your 4K TV

This is where I would usually say: leave it alone. Play the 1080p file on a TV that scales it. Modern TVs are good at this. If the picture looks soft, the cause is more likely the source (low bitrate, heavy compression) than the scaling.

Where upscaling in advance can help is when your TV’s scaler is poor, which is common on cheap displays, or when you want to add cleanup before scaling, such as denoise and deblock. In that case, process once with a good chain and keep the 4K result for that screen.

2. You are uploading to YouTube or another platform

This is the most common reason people upscale, and there is something real behind it. Video platforms apply different encoding settings to different resolutions. YouTube generally gives 4K uploads a higher bitrate and, for many viewers, the VP9 or AV1 codec, which compresses more efficiently than the H.264 used for lower resolutions on many videos. So an upscaled 4K upload can come out looking cleaner than the same footage uploaded in 1080p.

Creators who discovered this call it a workaround, and it is. It does not add detail. It reduces the compression damage that the platform adds. The cost is a much bigger file to upload and slower processing on YouTube’s side.

If your 1080p uploads look blocky after YouTube processes them, especially in busy scenes, try an upscaled 4K upload and compare. Make sure your export bitrate is generous. I cover the export side in best export settings for YouTube.

3. A client or a spec demands 4K

Then give them 4K, and be honest about the source in the delivery notes. A 4K file made from 720p is a legitimate deliverable if the spec says 4K, as long as nobody is being told it is native footage. Use a good filter, add a light sharpen, and encode at a proper 4K bitrate.

Settings I would use

If you want to do it yourself with FFmpeg, here is a starting point for a 1080p source:

ffmpeg -i input.mp4 \
  -vf "hqdn3d=2:1.5:3:2,scale=3840:2160:flags=lanczos+accurate_rnd,cas=0.3" \
  -c:v libx264 -crf 17 -preset slow -pix_fmt yuv420p \
  -c:a copy -movflags +faststart output_4k.mp4

Notes:

• The denoise is light. If the source is clean, drop it.
• scale=3840:2160 assumes 16:9. For other shapes, use scale=-2:2160.
• CRF 17 produces a big file. For 4K, that can mean 40 to 80 Mbps for busy footage, which is fine for a master and too much for web delivery.
• For web delivery, a CRF around 20 to 22 is a more reasonable balance.
• cas=0.3 is a gentle sharpen. At 4K, you can go lower than for 1080p because the extra pixels already feel crisper.

For a hardware encoder, the equivalent is HEVC on your GPU at 35 to 50 Mbps for a 30 fps 4K video. The file will be bigger than software x265 at the same quality, but it will finish in minutes.

How big will the file be?

People are often surprised. A rough guide for 4K at 30 fps:

Quality target	Typical bitrate	Size per minute
Web delivery, H.264	20 to 35 Mbps	150 to 260 MB
High quality H.264	40 to 60 Mbps	300 to 450 MB
Web delivery, H.265	12 to 25 Mbps	90 to 190 MB
Master quality	80 to 150 Mbps	600 MB to 1.1 GB

A ten minute 4K upload at 45 Mbps is around 3.4 GB. Make sure your connection can handle that, and check the platform’s file limits.

What about using AI?

AI upscalers claim 4K from 1080p as a headline feature. On the right material, such as faces and clean lines, the results can look sharper than a plain filter. They can also invent texture and flicker. I wrote a fuller comparison in AI video upscaler versus plain Lanczos. The short version: use AI on short, important clips, and keep the traditional chain for everything else.

Whichever you choose, process the video at 1080p or lower first for cleanup, then scale to 4K at the end. It is cheaper to denoise two million pixels than eight million, and the filters behave better at the original resolution.

Do not upscale from a damaged source

If your source has obvious blocks, the 4K version will have bigger, more obvious blocks. Spend the effort on cleanup first. The pixelated video article explains what causes blocks and which are fixable.

On this site, the full-quality render goes up to 1080p on the short side. That is a deliberate choice: for the kind of footage people bring us, which is mostly small and compressed, going beyond 1080p adds file size and cost without adding anything you can see. If you need 4K for a specific platform, you can take our 1080p result and scale it with the command above.

Frame rate and color

Two details come up when people upscale to 4K for platform reasons.

Frame rate. Keep it the same as the source. Converting 24 or 30 fps to 60 fps with interpolation produces smoothness, but also artifacts around fast motion. If the original was 30 fps, deliver 30 fps.

Color. 4K files are often tagged as Rec. 709 or Rec. 2020. Make sure the color tags match the source. Mismatched tags can make colors look washed out or oversaturated on some players. FFmpeg usually copies tags from the source if you do not change them. If your player shows a different look than your editor, this is the first thing to check.

When 4K is a bad idea

There are cases where upscaling will make things worse in practice.

Phones and old laptops. A 4K H.264 file at 45 Mbps may stutter on weak hardware. A 1080p copy plays smoothly everywhere.

Slow connections. Sharing a 3 GB file is not friendly. Offer a 1080p copy with the 4K one.

Archival. Do not replace your original file with an upscaled one. Keep the original. You can always upscale again later, and filters will be better in five years.

Streaming from your own server. Bandwidth costs add up. Unless viewers really use 4K, you will pay for pixels nobody can distinguish.

How to do it in the common editors

Not everyone wants a command line. Here is how the same job looks in the programs people use.

DaVinci Resolve. Set the timeline resolution to 3840 by 2160 in the project settings. In the same panel, under Image Scaling, set the “Mismatched resolution files” option to “Scale entire image to fit” for a clean fill or “Stretch frame to all corners” if you need to fill the frame. Then pick “Sharper” or “Smoother” for the resize filter. Sharper uses a wider kernel, similar to Lanczos, and Smoother is closer to bilinear. On the Deliver page, choose H.264 or H.265, set the quality to a fixed bitrate of 40000 kb/s or more, and render.

Premiere Pro. Create a sequence at 3840 by 2160. Drop your 1080p clips in and choose “Set to Frame Size” on the clips so that they scale with the sequence. Avoid the “Scale to Frame Size” option unless you want the scale to be baked into the clip’s scale property and keep its source resolution record. In the export settings, turn on “Use Maximum Render Quality”, set the bitrate to a target of 40 Mbps and a maximum of 60, and use two-pass VBR if you can afford the time.

Final Cut Pro. Make a new project at 4K. The app scales clips to fit by default. For export, use the “Master File” or a custom H.264 or HEVC setting rather than a preset that targets a small file.

CapCut and mobile editors. Pick the 4K export, the highest bitrate and the HEVC codec if your destination supports it. Be aware that many mobile apps use a fairly simple scaler, so the result may look softer than the same job in a desktop editor.

None of these will give you detail you did not film. They all do what the FFmpeg command does, with a different interface.

What YouTube actually recommends

For uploads, YouTube publishes recommended bitrates. For SDR content at standard frame rates, they are in the neighborhood of 8 Mbps for 1080p and 35 to 45 Mbps for 4K. At 60 fps, the numbers are higher: about 12 Mbps for 1080p and 53 to 68 Mbps for 4K. These are recommendations for the upload file, since YouTube re-encodes it anyway.

That is why people who want cleaner YouTube video go for a high bitrate in the upload. You are giving YouTube’s encoder the best possible starting material. And if you upload in 4K, the higher resolution tier gets treated better in the re-encode. Both effects help, and neither requires the picture to contain real 4K detail.

If you try this approach, test it. Upload the same 1080p footage twice, once as 1080p and once as an upscaled 4K, set both to private, wait for the processing to finish, and compare them at full screen on the same device. Do not trust the first few minutes after upload: YouTube often shows a low quality version first, and the higher tiers appear later.

A test you can run in ten minutes

Here is a quick method to decide whether upscaling is worth it for your own footage.

1. Pick a 30 second section with a mix of still and busy shots.
2. Export it as 1080p at a good bitrate.
3. Export the same section as a 4K upscale with Lanczos and a mild sharpen.
4. Play both on the screen where the video will mainly be watched.
5. Sit at your normal distance. Ask someone else to pick which is which without telling them.

If neither you nor your helper can reliably tell, skip the upscale. If the 4K looks cleaner in the busy sections, the platform or the screen is rewarding the higher resolution, and it is worth doing for that destination.

I find a blind test humbling. It is easy to see differences when you know which file you are looking at and much harder when you do not.

Common mistakes with 4K upscales

Using a low bitrate. Upscaled 4K at 10 Mbps looks worse than the 1080p original at 10 Mbps, because the encoder has more pixels to describe with the same bits. If you upscale, raise the bitrate in proportion.

Sharpening too much. Over-sharpening on a clean 4K canvas makes edges glow. A mild sharpen at the end is enough.

Forgetting the audio. Copy the audio stream rather than re-encoding it, unless you have a reason to change it.

Mixing resolutions in one timeline. If you have 4K, 1080p and 720p clips in one project, the lower resolution clips look soft next to the sharp ones. Either upscale them with the same filter before editing, or accept the difference.

Leaving the source open for the next job. Keep the original files. If you ever need to redo the upscale with a better tool, you will be glad you did.

Summary

Upscaling to 4K makes sense for three reasons: a platform that treats 4K uploads better, a delivery spec that requires it, or a cleanup chain you want to apply once. It does not make footage truly 4K, and it is not worth doing just to feel better about a soft picture. A clean source, a generous bitrate and a good filter are the things that matter. Check the result on the screen that will be used, preferably blind, and keep the original safe.

My recommendation

If you are curious, try this. Take a one minute segment of your 1080p video. Make a 4K upscale with Lanczos and a light sharpen. Play both files on your 4K screen from the same distance you normally sit. Most people find the difference is tiny, and sometimes only visible in menus and titles.

If the difference matters, for platform reasons or a spec, go ahead. If it does not, keep the 1080p and spend your time on a better source, a better bitrate, or a better camera setting. A clean, well-compressed 1080p video beats a mediocre 4K upscale nearly every time.

If your video is small or blocky to begin with, try the cleanup first. You can upload a short clip here to see what deblocking, denoising and scaling do to it before you decide on a final size. Seeing your own footage beat any article, mine included.