Face-Aware Portrait Crop for Vertical Video

Face-Aware Portrait Crop for Vertical Video

Cropping a landscape video to portrait is easy if you do not care what gets cut off. It is harder when the speaker’s face matters, the captions need room, and the same recording has to work on Shorts, Reels, and TikTok.

The crop box needs intent. Center crop is fine for a perfectly centered speaker. Most real footage is not that polite. Velocaption’s face-aware crop starts from the detected face, adds breathing room, then fits that padded box into the selected aspect ratio.

Key Takeaways

• Use 9:16 for vertical social video.
• Face-aware cropping needs padding around the detected face.
• Velocaption uses Apple’s Vision face rectangles locally on Mac.
• Keep landscape and portrait layouts separate so captions stay readable.

Why Does Center Crop Fail?

A 16:9 frame cropped to 9:16 keeps only the middle slice of the original image. If the speaker leans, sits off-center, or shares the frame with screen content, a blind center crop cuts away useful context.

This is why “convert to vertical” tools often produce awkward results. They solve the aspect ratio but not the composition.

Good portrait conversion needs three decisions:

• what subject matters,
• how much headroom to keep,
• where captions and overlays will go.

If all you need is a quick browser crop, use the free video cropper. If you need captions, face framing, and separate portrait layout, keep it in the desktop editor.

How Does Face-Aware Cropping Work?

Face-aware cropping detects the face rectangle, finds its center, adds padding around it, then fits the requested aspect ratio around the padded area. The face is the anchor, not the final crop.

Velocaption’s crop math uses concrete multipliers:

That vertical bias is small but useful. A detected face box usually covers the face, not the composition. Centering exactly on the face can put the chin and shoulders too low. Biasing slightly downward gives the portrait crop a more natural speaker frame.

Under the hood, the app picks the best face by area and confidence, then clamps the crop to the source bounds. If the padded box is too large, it scales down instead of spilling outside the frame.

Why Use Apple’s Vision Framework?

Apple’s Vision framework exposes face rectangle detection through VNDetectFaceRectanglesRequest. Velocaption calls that local macOS API from the Tauri backend, receives normalized face boxes, and never needs to upload a frame for face analysis.

The important part is locality. A crop assistant should not require sending a client recording or private webcam frame to a server. Face detection can be a local utility.

Velocaption’s bridge converts Apple’s bottom-left image coordinate system into the top-left coordinate system used by the crop canvas. Then the React crop UI applies the calculated rectangle.

If no face is found, the app does not pretend. It surfaces “No face found in this frame” and leaves the crop under manual control.

What Aspect Ratio Should You Pick?

Use 9:16 for vertical social output. Use 1:1 when the destination is square. Use 16:9 when you are preserving the original landscape presentation.

Velocaption exposes these crop presets:

The “Canvas” option is useful inside a project because it follows the current export orientation. If the project is portrait, Canvas resolves to 9:16. If it is landscape, Canvas resolves to 16:9.

For a full workflow, read Convert Landscape Video to Portrait with Captions.

Where Should Captions Go in Portrait?

Portrait video has less horizontal room, and the face often competes with captions. Do not blindly reuse the landscape caption position. Make portrait a separate layout.

Velocaption lets the transcript stay shared while the spatial layout changes. That means the word edits, confidence scores, split/merge fixes, and caption style survive when you switch formats. Only the canvas arrangement changes.

In practice:

• Put captions higher than the bottom UI safe area.
• Keep the face bubble away from the caption line.
• Use shorter caption phrases for vertical output.
• Use a background fill when the source frame does not cover the portrait canvas.

The floating face guide covers per-scene position, crop, roundness, and size. The caption styles guide covers reusable typography.

When Is a Portrait Box Better Than a Crop?

Cropping removes pixels. A portrait box keeps the whole source visible and places it on a 9:16 canvas with a background. Use it when the screen content matters more than filling the frame.

For tutorials, product demos, and course videos, a hard 9:16 crop can remove the UI you are explaining. A portrait box keeps the landscape video intact, adds top and bottom space, and gives you room for captions, face, or a title.

Velocaption has a batch portrait box utility for exactly that job. It can center mixed-orientation videos on a 9:16 canvas with solid, gradient, or blurred backgrounds.

The choice is simple:

• Crop when the speaker is the subject.
• Portrait box when the screen content is the subject.
• Use both when a course clip needs a face bubble plus readable screen content.

How Velocaption Turns One Recording Into Two Formats

The best workflow is not “export landscape, re-import, crop, then fix captions again.” That wastes the transcript.

Velocaption keeps one transcript and separate layouts:

1. Generate local captions once.
2. Review words and timing once.
3. Configure the landscape layout.
4. Switch to portrait.
5. Reframe face, captions, background, and overlays.
6. Export each format separately.

That gives you a YouTube version and a Shorts/Reels version from the same source edit. If the video also needs compression or conversion afterward, the free browser tools can handle one-off file cleanup.

Frequently Asked Questions