How to crop to the subject of each shot when it is not a face

Last updated 2026-09-10 · 8 min read

What is the short version?

A vertical crop has to choose what to keep, and by default it keeps the face. In a product video that means the demonstration shots come back cropped from the top, with the item you are demonstrating half out of frame.

1. Go to Vizard Agent with the cut.
2. Say which shots are about the object.
3. Ask for those crops to be re-anchored to it.

What do you need before you start?

The footage and a division of the shots. Vizard Agent needs to know which shots are person-led and which are object-led, because that distinction is editorial — the same wide shot can be about either depending on what the line over it is saying.

• The clips. Or the assembled cut.
• Which shots feature the object. Named by content.
• What must stay whole. The whole item, or a detail.
• Whether the face still matters there. Sometimes both.
• The target format. Vertical is where this bites.

What do you type into Vizard Agent?

Split the shots by subject inside the instruction itself. Asking Vizard Agent for a better crop gets you an adjustment to the existing one; telling it which shots are about what changes the rule the crop is following in the first place.

Prompt

Variants worth knowing:

• "Where the bag is the subject." Names the group.
• "Fully in frame." Not just present.
• "Keep following her elsewhere." Protects the rest.

What does Vizard Agent actually do?

Here is the sequence Vizard Agent worked through on a real product reel reframed to vertical. Its first pass was face-led throughout, which was exactly right for half the film and quietly wrong for the other half of it.

1. Reframes the cut and reviews the result.
2. Identifies the shots where the object is the subject.
3. Maps the object's position in each source shot.
4. Measures where it actually sits, rather than estimating.
5. Reads off those positions as numbers.
6. Re-crops those shots anchored to the object.
7. Checks the re-anchored framing on real frames.
8. Leaves the face-led shots untouched.
9. Confirms the two groups still cut together.

Step three is what makes this per-shot rather than global. An object does not sit in the same place in every shot — it is held up in one, on a table in another — so one anchor point cannot serve them all and each shot needs its own measurement.

Step eight is worth stating because it is easy to lose. A rule applied to the whole film to fix half of it breaks the other half, and the point of separating the groups is that each keeps the anchor that suits it.

Step nine catches the side effect. Two differently anchored groups can cut together awkwardly if the subject jumps position across the join, so the assembled result gets checked rather than each group on its own.

What does the result look like?

A vertical cut where the demonstration shots show the whole item and the talking shots still hold the person. Nothing is cropped out of the shots that exist to show it, which is usually the reason the video is being made.

Vizard Agent tells you which shots it re-anchored, so a shot you disagree about is one adjustment rather than a re-run.

When does this not work well?

Some frames cannot hold both. If the person and the product are at opposite edges of a wide shot, a vertical crop has to choose, and the answer is either a different shot or a layout that shows both.

• Person and object far apart. The crop must choose.
• An object that moves across the frame. It needs tracking.
• Very wide shots. Cropping to the object loses the context.
• Several objects. Which one is the subject?
• Shots that are about both. Consider a split layout.

How do you fix a result that came back wrong?

Name the shot and say what it is actually about. Because Vizard Agent grouped the shots by subject rather than cropping them individually, a note like that usually means moving one shot from one group into the other rather than adjusting a crop by hand.

• "This one is about the bag too." Moved to that group.
• "The bag is still cut off." Re-measured on that shot.
• "Now her head is cropped." Both matter; a wider anchor.
• "It jumps at the join." The two anchors get reconciled.

How does Vizard Agent compare to doing it yourself?

By hand automatic reframing is run over everything, because it is one setting, and the product shots are fixed afterwards one at a time by dragging. It works and it takes an afternoon, and the anchors end up inconsistent between similar shots.

Common questions

Why does it default to faces? Because most vertical reframing is for talking video, where the face is the subject.

Can it tell which shots are which? Often, but Vizard Agent will ask when the shot could be either.

What if the object moves? Then Vizard Agent tracks it rather than using a fixed anchor.

Can both stay in frame? Sometimes. Vizard Agent will say when the frame cannot hold both.

Does it work on a finished cut? Yes, and Vizard Agent re-renders only the shots whose crop changes.

What about a detail rather than the whole item? Tell Vizard Agent which, and it anchors to that instead.

Will the shots match each other? Vizard Agent checks the groups cut together, not just individually.

Can I give it the positions? Yes, though Vizard Agent will measure them on a frame to confirm.

Does this apply to square as well? Yes. Vizard Agent applies it wherever the crop has to choose what to keep.

What if there are two products? Tell Vizard Agent which one is the subject of each shot.

Can it use a split layout instead? Yes. Vizard Agent will suggest one when a shot genuinely needs both in frame.

Will it crop into the product? Not if it is anchored to keep it whole, which is the instruction.

Does the line over the shot matter? Considerably. What is being said usually decides the subject.

Why did my reframe cut the product off? Because it was following the face, which is a reasonable default and the wrong one here.