The classic face shapes

Most guides use six or seven categories: oval, round, square, rectangle (oblong), heart, diamond, and sometimes triangle (pear). They're defined by a handful of ratios:

• Face length vs. width (long faces are oblong/rectangle, short and wide ones are round/square)
• Forehead width vs. cheekbone width vs. jaw width (wide forehead and narrow chin points to heart; widest at the cheekbones points to diamond)
• Jawline angle (sharp, angular jaws read as square/rectangle; soft curves read as round/oval)

What the AI actually does

1. Face detection finds the face in the image and crops it.
2. Landmark detection places dozens of points along the jaw, chin, cheekbones, brows and hairline.
3. Geometry turns those points into ratios — length/width, forehead/jaw, jaw angle.
4. Classification maps the ratios (often with a trained model rather than fixed thresholds) to the closest face shape, usually with a confidence level, because real faces are frequently a mix of two shapes.

Why two tools can give different answers

• Head tilt and camera distance. Phone wide-angle lenses close up make the middle of the face look larger; tilting the chin up shortens the jaw.
• Hair covering the hairline or jaw. The model can't measure what it can't see.
• Expression. A big smile widens the cheeks and changes the jaw line.
• Borderline faces. If you're between oval and oblong, small changes flip the label — which is normal.

How to take a photo that gives an accurate result

• Face the camera straight on, eyes level, neutral expression.
• Hold the phone at arm's length or further (or use the 2x lens) to reduce distortion.
• Pull hair back so the forehead and jawline are visible.
• Use even, soft light — a window in front of you, not overhead light.
• Skip beauty filters and heavy retouching; they reshape the jaw and skin.
• Try 3–5 similar photos and look at the most common result.

Trying it yourself

I tested this with FaceAnalysis AI, which gives a free face shape estimate together with a 1–10 photo score from one clear, front-facing photo (you can add up to five photos of the same person for a steadier estimate). There's also a dedicated face shape analyzer page. The tool is upfront that the score is an AI estimate based on patterns in its training data, not a scientific measurement of anyone's worth — which is the right way to treat any of these tools. If you want more, it offers an optional paid full report with hairstyle, skincare and wardrobe suggestions.

Using your face shape

• Haircuts: add height for round faces, add width at the sides for long faces, soften angles for square faces.
• Glasses: choose frames that contrast your shape — rounder frames for angular faces, angular frames for round faces.
• Beards: a fuller chin lengthens a round face; shorter sides keep a long face balanced.

Face shape is a starting point, not a rule. Use it to narrow your options, then choose what you actually like.

The main metadata containers

EXIF is the oldest and most common. Cameras and phones write it into JPG (and often HEIC/WebP) files: camera make and model, lens, exposure, date and time, orientation, and — if location services are on — latitude and longitude.

XMP (Extensible Metadata Platform) is an XML block that editing software uses for ratings, keywords, edit history, and tool names. Many generative tools write fields here that identify how an image was made.

IPTC fields are used by newsrooms and stock libraries for captions, credits, copyright notices, and keywords.

C2PA content credentials are a newer, cryptographically signed manifest that records where an image came from and what was done to it (for example "created with an AI model" or "edited in app X"). Platforms can read this manifest to show a provenance label.

PNG text chunks (tEXt, iTXt, zTXt) can store arbitrary key/value text. Some AI image tools store the full generation prompt and settings here.

How to check what a file contains

• On the command line, exiftool photo.jpg prints nearly everything, including XMP and IPTC.
• exiftool -G1 -a -s photo.jpg groups the output by container so you can see which block each field lives in.
• For C2PA, the open-source c2patool photo.jpg shows the manifest if one exists, or you can drop the file into the Content Credentials "Verify" site.

Why "remove EXIF" is often not enough

Many "remove EXIF" options in phones and websites only delete the EXIF block. XMP, IPTC, C2PA manifests and PNG text chunks can survive. The most reliable approach is to re-encode the image into a brand-new file that only contains the decoded pixels — nothing from the original container gets copied.

You can do that yourself:

# re-encode with ImageMagick and drop all profiles/metadata
magick input.jpg -strip output.jpg

# or with exiftool, delete every writable tag
exiftool -all= -o output.jpg input.jpg

A no-install option that keeps files on your device

If you don't want to install tools or upload private photos to a server, a browser-based cleaner can do the re-encode locally. I use the free Remove AI Label / metadata cleaner from SynthClear: it decodes each image in your browser and writes a fresh copy without the original EXIF (including GPS), XMP, IPTC, C2PA content credentials, or PNG text chunks. Nothing is uploaded; it handles JPG, PNG and WebP up to 15 MB each, 20 images at a time. JPEGs are recompressed at high quality and WebP comes out as PNG.

What metadata removal does not do

Stripping metadata only changes the file container. Pixel-level watermarks stay in the image. Google's SynthID, for example, is embedded in the pixels themselves and is designed to survive re-encoding, cropping and compression, so it can still be detected after the metadata is gone. SynthClear also offers a separate paid SynthID tool on its homepage, but that's a different job from metadata cleaning.

Use it responsibly

Metadata cleaning is great for privacy — removing GPS from photos of your house, or camera serials from images you publish. Only process images you own or are authorized to edit, and keep following the disclosure rules of the platform, client, or publication you're posting to. If content is AI-generated and a platform requires you to label it, removing a file's metadata doesn't change that obligation.

Quick checklist before you share a photo

1. Run exiftool (or any metadata viewer) to see what's inside.
2. Remove location and personal data by re-encoding to a fresh file.
3. Re-check the output to confirm the fields are gone.
4. Add any required disclosure yourself in the post text.