How to Take the Perfect Photo for a Resemblance Test
AI facial recognition can return surprisingly different results depending on the photos used. Here are 7 simple rules to maximise the accuracy of your resemblance test — and 3 traps to absolutely avoid.
Why photo quality changes everything
Our Look Like Me LLM Engine (and any facial recognition engine) extracts a mathematical "vector" from the face out of the image. If the image is blurry, poorly lit, or the face is partly hidden, the vector is less precise — and the resemblance score can vary 10-30% from the "true" value.
Good news: with a few simple rules, you can shoot a perfect photo in 30 seconds with any modern smartphone.
The 7 golden rules
1. Face front, not in profile. The AI is mainly trained on frontal faces. Profile or three-quarter angles give less reliable results.
2. Natural light, ideally diffused. Near a window on an overcast day, or outdoors in shade. Avoid backlight and direct flashes.
3. Neutral expression, mouth closed. A smile or forced laugh distorts the jaw and cheeks. Relaxed face, eyes straight to the lens.
4. No glasses (unless the person has worn them since childhood). Glasses obstruct the eyes and brow area.
5. Hair off the face. Strands covering the forehead or cheeks reduce usable facial landmarks.
6. Optimal distance: 50-80 cm. The face should fill about 50-70% of the image height. Too far = not enough pixels, too close = distortion.
7. Minimum resolution 800x800 pixels. All modern smartphones do this without issue. Avoid WhatsApp-compressed photos (use the original or "original quality" option).
Discover who you really look like
Try the Family Plan — €4.99The 3 absolute traps to avoid
Trap 1: group photo. If you crop a face out of a family photo, the resulting resolution is often too low. Take a dedicated photo, even quickly.
Trap 2: heavy make-up or thick beard. Make-up hides key facial landmarks (brows, eye lines). A full beard masks the jaw and mouth contour.
Trap 3: very old, poorly digitised photo. A paper photo scanned at low quality often has artefacts that confuse the AI. To test with an old photo, shoot a smartphone photo of the paper print in good light, rather than a low-resolution scan.
How to test with a newborn or young child
Babies are a special case. Specific recommendations:
- Awake and calm. Not while crying or laughing — face too distorted.
- Hold the head straight, not tilted — tricky with a newborn, but try.
- Natural light across the whole face, no harsh shadow on one side.
- Under 6 months, take results with a grain of salt. The baby's face is still very weakly differentiated — the AI may hesitate.
Tip: take several photos in sequence and run the test on each. If scores diverge a lot, your photos aren't stable enough. If they're consistent, you have your answer.
When to use old photos
Testing a child against an old photo of their grandparent at the same age is one of the most moving experiences AI enables. It's also the riskiest technically.
To maximise the chances:
- Take a photo of the old paper photo (laid flat, diffused light), rather than scanning
- Avoid damaged photos (folds, scratches, extreme yellowing)
- Prefer frontal photos — black-and-white 50s portraits often beat blurry 70s polaroids
The engine can hit real emotional sweet spots: an 80% score between your child and your grandmother at 6 is a family experience few tools can offer.