Zooming in without going soft

You are watching the footage back and the important part is too small. The hands, the stitch, the label, the thing you were talking about, sitting in the middle of a wide shot of a room.

So you scale the clip up in the edit, frame it nicely, and it looks worse. Not dramatically. Just softer, mushier around the edges, slightly out of keeping with the shots either side of it, in a way that is hard to point at and impossible to unsee.

A crop gives you fewer recorded pixels to work with. Source quality, scaling and export settings also affect how soft the result looks, so check the full chain before blaming the camera.

What is a punch-in doing?

Cropping, and then stretching what is left to fill the frame again.

A camera lens zooming in gathers new detail: the light coming through the glass changes and the sensor records more information about a smaller area. A punch-in in the edit does nothing of the sort. It selects a rectangle from the picture you already have and enlarges it, and there is no mechanism by which the software can know what was between two pixels. It guesses, competently, and guessing looks like softness.

So the useful way to think about it is a budget. The recording has a fixed amount of detail in it. Every punch-in spends some. When it is gone, it is gone, and nothing bought afterwards brings it back.

The pixel budget

This part is straightforward arithmetic and you can do it before you next press record.

A UHD frame at 3840 × 2160 has twice the width and height of 1920 × 1080. At the same aspect ratio, a crop using half its width and height still contains a 1920 × 1080 raster. That is pixel headroom, not a guarantee of sharp detail.

Half the width is a much bigger move than it sounds. It doubles the apparent size of everything in frame. A wide shot of a person at a workbench becomes a shot of that person from the waist up. A mid shot becomes a close-up of two hands. For most of what people want from a punch-in, that is the entire requirement met, for nothing.

Record 1080p and deliver 1080p and you have no budget at all. The first percentage you crop is already being invented. It can still be worth doing for a small, brief move, but you are spending from zero.

One caution, because resolution is a ceiling. Soft focus, a smeared lens, poor light or heavy compression all mean the extra pixels do not carry extra detail, and a 4K recording of an out-of-focus subject punches in exactly as badly as a 1080p one. The number is permission, not quality.

Early signs of softness

Fine texture goes first. Hair, fabric weave, the grain in wood, small printed text on a label. These turn waxy well before anything looks pixelated, which is why people miss it on the small preview window and then see it on a television.

The second sign is movement. A punched-in shot with motion in it looks softer than a static one, because any motion blur, camera shake and compression already baked into the frame get enlarged along with the picture. If you must push past what the budget allows, do it on a still moment and hold it briefly.

Where should the close-up be decided?

At the camera, and that is the only useful thing in this post.

Every option in the edit is a way of managing a shortage. Every option at the shoot removes the shortage. Deciding you will want a close-up before you film costs nothing at all, and it is free whether you solve it by recording above your delivery size or by simply filming a close-up.

Where somebody is teaching with their hands, the detail angle gets recorded above the delivery size for exactly this reason, so the hands can be tracked and punched into later without the picture softening. Nobody has to buy anything for that. It is a decision taken before anybody opens an edit, in a conversation about what the videos are for.

What if the footage is already shot?

Then face the situation, because this is where people spend money badly.

Software that promises to invent detail exists and it is improving, but it is a poor first purchase for somebody who films weekly and could simply record a close-up next time. If you are tempted, the question to answer is whether you are rescuing one important video or building a habit. Rescuing one is fine. Building a habit is paying every month for a problem that a note on your shoot list would remove.

The free options are better than they sound:

1. Compare a direct cut with a moving punch-in. 2. Check the crop beside neighbouring shots at the intended output size. 3. Watch how motion blur and shake change when enlarged. 4. If the detail is not usable, consider another shot, a diagram or a clear verbal explanation.

Before the shoot, name the detail the viewer needs to see. Film a dedicated close-up or record with enough usable resolution for the planned crop, depending on the equipment and action.

Sources and evidence

What here is established resolution arithmetic and what is our own method.

What is established. Recommendation ITU-R BT.709 defines the standard 1920 by 1080 high-definition frame, and Recommendation ITU-R BT.2020 defines the standard 3840 by 2160 ultra-high-definition frame, exactly four times the pixel count. That is where the half-width, four-times-area figures in this piece come from: they describe the formats, not a house estimate. Neither standard says anything about what happens when software enlarges a crop taken from inside those pixels. That a punch-in interpolates rather than recovers real detail is how digital enlargement works in general, not a claim either recommendation makes itself.

Our method. Deciding the close-up before the shoot, and recording the detail angle above the delivery size so nobody has to invent it later, is the practice here.

References.

• ITU-R. Recommendation BT.709-6: HDTV parameter values. itu.int . In-force HDTV standard; picture raster 1920x1080 (with BT.2020 gives the four-times-pixel crop arithmetic).
• International Telecommunication Union (ITU-R). Recommendation ITU-R BT.2020: Parameter values for ultra-high-definition television systems. itu.int . Defines the 3840x2160 UHDTV-1 pixel format that, with BT.709's 1920x1080, grounds the 'four times the pixel count' crop-headroom arithmetic.