Smart Face Crop

Why face-aware crops beat manual crops

Cropping a portrait manually is harder than it looks. The right framing depends on where the face is in the source photo, which way it's tilted, how much shoulder is visible. Most people end up with the face dead-centre — which looks fine but reads as ID photo, not portrait.

Professional photographers use the rule of thirds: imagine the photo divided into 3×3 grid; place the eyes on the upper horizontal line. This gives the photo natural visual weight, leaves breathing room above the head, and looks "photographed" rather than "cropped from a phone shot".

What this tool actually does

On upload, the photo goes to our private worker. InsightFace runs face detection and returns the bounding box of the largest face. The Python code:

1. Computes the crop dimensions based on the target aspect ratio + source dimensions.
2. Places the crop centred horizontally on the face, anchored vertically so the face sits at 42% from the top (rule of thirds + a small adjustment for headshot proportions).
3. Clamps the crop to the source bounds — never crops outside the photo.
4. Returns a JPG at quality 0.94.

Total time: ~400 ms warm, ~3 s cold (InsightFace model load). Memory: ~1 GB across the worker process.

Use cases

• LinkedIn / corporate avatar (1:1). Crop from any portrait to a square, face properly placed.
• Instagram portrait posts (4:5). The most engaging Instagram aspect for portraits, often hard to nail manually.
• Resume photos (3:4). Pair with the Resume Photo Sizer for the country-specific mm dimensions.
• YouTube thumbnails / Twitter banners (16:9). Wide crops of portrait photos are notoriously hard manually — face-aware framing fixes it.

Related tools

Use the Profile Picture Resizer for manual platform-specific framing with a circle preview. The Smart Crop is the "I trust the algorithm" version. For full control over the crop position, use the framing editor with manual drag + zoom.