Does the camera matter? Phone vs. DSLR?

Almost not at all. Any phone made after 2020 has more than enough sensor for AI styling. What matters is light, angle, and stillness.

No. A neutral closed-mouth expression gives the model the cleanest jaw and lip line. Save the smile for after the haircut, not before.

Workable, but worse than asking someone else to take it. Selfies tend to distort proportions because the lens is too close. Hand the phone to a friend and step back four feet.

How to Photograph Your Face for AI Styling (Without Looking Like a Mugshot)

The single biggest thing that separates a useful AI hairstyle preview from a useless one is the input photo. Not the AI model. Not the prompt. The photo.

I've spent the last year testing this — first as part of a research project at Adobe, more recently for Style Hem's internal photo-quality work — and the gap between a good input photo and a typical phone selfie is enormous.[3] Here's what actually moves the needle.

The four rules, in order of importance

1. Soft, even, front-facing light

This one matters more than the other three combined. AI models reconstruct your face by reading light and shadow; harsh shadows get baked into the output as features.[2]

The free, perfect light source is a north-facing window at midday, with you facing it. No filter, no ring light, no fancy setup.

What to avoid:

Overhead light (kitchen, bathroom, office). Casts shadows under the eyes and chin that the model interprets as bags or jowls.
Backlight (window behind you). Underexposes your face; AI fills in detail with averages, and you stop looking like yourself.
Mixed color temperature (warm bulb + daylight). Adds color casts the model can't always neutralize.

2. Frame from the collarbone up

Not just the head. Not from the chest. Collarbone.

There's a practical reason: most current image models — including the GPT Image 2 model that powers a lot of styling tools — were trained on portraits framed this way, and their hair-edge detection is most accurate at this crop.[1] Tighter crops lose context; wider crops dilute the resolution where it matters.

3. Neutral expression, eyes on the lens

A small smile is fine. A teeth-showing smile causes problems — it changes your jaw shape, your cheek volume, and the tension along your hairline. The AI ends up trying to reproduce all of that in the output, and the output looks like a slightly different person.

Mouth closed, eyes on the camera, head straight (not tilted). It's not a mugshot — your shoulders can be relaxed, you can look like a normal human — but the face itself should be at rest.

4. Resolution: 1024 px short-side is the sweet spot

This surprises people. Bigger is not better.

1024pixels short-side, the sweet spot^[1]