How to get a close-up of one person out of a two-person wide shot
Tell Vizard Agent which person you want alone and during which shots. It finds every window where that person is speaking on the wide camera, crops each one to a natural close-up without stretching, then rejoins the master frame by frame and checks the runtime and the audio track came out identical to the approved cut.
What is the short version?
A two-camera interview usually has one camera pointed at both people, so there is no close-up of the interviewer at all. There does not need to be. The wide shot contains one, at a lower resolution, if you crop it only while they are talking.
- Go to Vizard Agent with the approved edit.
- Say which person to isolate and on which camera.
- Say nothing else in the edit may change.
What do you need before you start?
You need the wide shot at its original resolution and a clear description of the framing you want. Framing described in body terms — upper chest, head room — travels better than a percentage, because it means the same thing at every point in the shot as people lean and move.
- The approved edit. The version you already signed off.
- The original wide-camera file. Cropping needs the pixels.
- Who to isolate, and on which shots.
- How tight. "From the upper chest up" is a good instruction.
- What must not move. Usually everything else.
What do you type into Vizard Agent?
Two sentences do the work: the framing, and the freeze on everything else. The second one matters more than it looks, because a request to change one shot is easily read as permission to improve the neighbouring ones as well.
Prompt
Variants worth knowing:
- "Not stretched or distorted." Rules out the lazy fix.
- "Only during her sections." Not the whole shot.
- "Change nothing else." The instruction that protects the edit.
What does Vizard Agent actually do?
Here is the order Vizard Agent worked in on a real two-camera documentary interview where the interviewer had no close-up of her own. Most of the list is not cropping — it is making sure the corrected master matches the approved one everywhere the crop does not apply.
- Identifies the speaker's exact speaking windows in the wide source.
- Maps each window onto the wide-camera portions of the approved edit.
- Looks at the frames you flagged to confirm what is actually there.
- Builds a crop for those windows only.
- Renders a test at the timestamp you named before doing anything else.
- Reads the actual error when a render fails, rather than retrying blindly.
- Simplifies the processing graph until it renders reliably.
- Processes the timeline in windows, splitting any that fail.
- Measures frame counts at every window boundary to prevent drift.
- Trims each window to frame-accurate boundaries before joining.
- Aligns the joined master to the approved duration without re-encoding audio.
- Verifies the duration, the geometry and the untouched audio stream.
- Checks the named timestamp and every later section in the picture.
- Uploads the corrected master and plays it back to confirm.
Step nine is the part that separates this from a quick crop. Rebuilding a long edit in pieces introduces a frame here and a frame there, and a few frames of drift across a twenty-minute interview will pull the captions off the words at the end.
Step eleven is the promise being kept. The audio stream is copied rather than re-encoded, so the corrected master is provably the same soundtrack as the approved one — which is what "do not change the audio" has to mean in practice.
What does the result look like?
The same edit you approved, with one difference: during that person's answers on the wide camera, the frame is a natural close-up of them alone. The runtime is identical, the audio is the same file, and every other shot is exactly where it was.
The close-up is softer than a real second camera. In context, nobody notices.
When does this not work well?
Cropping trades resolution for framing, so the wide shot has to have pixels to spare. It also has to have the person reasonably placed in the frame, because a crop cannot move somebody towards the centre without cutting something else off.
- A low-resolution wide. The crop will be visibly soft.
- The subject at the frame edge. There is no headroom to keep.
- Constant movement. A fixed crop loses them; tracking is a different job.
- Both people in one gesture. Cropping cuts the exchange in half.
- Captions across the body. Say where they should sit instead.
How do you fix a result that came back wrong?
Give a timestamp and say what you see there. Vizard Agent keeps the speaking-window map and the crop it built for each one, so a note about one moment changes one window rather than sending the whole master round again.
- "Still the wide shot at 0:09." That window's crop rebuilt and verified there.
- "Too tight." The frame widened, headroom restored.
- "She looks stretched." The aspect corrected; crop, never scale.
- "The captions cover her." Moved clear of the new framing.
How does Vizard Agent compare to doing it yourself?
The manual route is either a re-edit, which risks everything that was already approved, or a crop layered on top, which drifts out of sync unless the frame counts are watched at every join. Both are slow on a long interview with many short answers.
| By hand | Vizard Agent | |
|---|---|---|
| Finding the windows | Scrubbing for her lines | From the speaking intervals |
| The crop | One setting for the shot | Per window, checked on frames |
| Timing after rebuilding | Drifts a frame at a time | Frame counts measured at every join |
| The audio | Re-encoded with the render | Copied, provably unchanged |
| The rest of the edit | At risk | Untouched by construction |
Common questions
Will it look soft? Softer than a real close-up. Vizard Agent crops from the highest resolution available.
Can it follow her if she moves? Yes, as tracking. Say so and Vizard Agent plans it that way.
Does the whole video get re-rendered? Only the affected windows. Vizard Agent rejoins them to the approved master.
How do I know the audio is unchanged? Vizard Agent copies the stream and verifies it afterwards.
Can it do both people? Yes. Ask Vizard Agent for a crop per speaker and it will alternate.
What about the captions? Vizard Agent keeps them clear of the new framing rather than re-timing them.
Will the runtime change? No. Vizard Agent normalises the join back to the approved duration.
Can I name the exact moment to check? Yes, and you should. Vizard Agent verifies that frame specifically.
What if the crop fails on one section? Vizard Agent splits it into smaller windows and renders those.
Does it work with a GoPro or an action camera? Yes. Wide lenses are the usual source for this.
Can it remove the background clutter too? Cropping removes most of it. Tell Vizard Agent what else bothers you.
What if I want a wider version as well? Ask Vizard Agent for both and compare them.
Will it tell me which windows it changed? Yes. Ask Vizard Agent for the list with timestamps.
Why not just zoom the finished video? Because zooming the finished file scales an already-compressed picture, while cropping the original keeps every pixel the camera recorded.