Getting an AI headshot that does not look AI

What it is good at

• Fast turnaround for LinkedIn, team pages and founder bios without booking a photographer.
• Easy to produce a consistent set across a whole team from one reference image each.
• Cheap to run many variations of expression and outfit before picking a final.
• Editing tools can fix a real photo rather than fabricating a face from nothing.

Where it falls short

• Skin and hair still read as too smooth unless you explicitly ask against it.
• Ears, glasses and hands near the face are the most common failure points.
• Reusing the same reference photo across tools produces subtly different faces each time.

How to use it for creative work

1. 01

   Start from a real photo, not a blank prompt

   Upload an actual photo of the person and ask the tool to edit lighting, background and outfit rather than generate a stranger. This keeps the bone structure and proportions correct, which text-only generation gets wrong most often.

2. 02

   Specify the lens and distance

   Ask for an 85mm or 105mm portrait lens at a comfortable working distance, not a wide lens close to the face. Wide lenses distort the nose and forehead in a way that reads as fake even when everything else is right.

3. 03

   Name the light quality, not just the mood

   Say 'soft north-facing window light' or 'large softbox at 45 degrees' rather than 'good lighting'. Flat, shadowless light is the single biggest tell of an AI headshot, so ask for a visible falloff and a gentle shadow under the jaw.

4. 04

   Demand texture explicitly

   Add 'visible skin texture, natural pores, slight under-eye shadow, no retouching' to the prompt. Without this instruction, most models default to smoothed, poreless skin that looks like a video game character.

5. 05

   Generate in batches and cull hard

   Produce eight to twelve variants per pose and reject anything with warped ears, doubled eyebrows or glassy eyes. Keep the two or three that would survive a colleague asking 'is that a filter?'

Ready-to-use prompts

Using the attached photo of [name], keep the face and features unchanged. Professional headshot, 85mm lens, soft large softbox at 45 degrees camera left, plain [colour] backdrop, visible skin texture and natural pores, no retouching, slight natural asymmetry, confident relaxed expression.

Same subject as attached, unchanged face. Half-body portrait near a large window, soft diffused daylight from the left, shallow depth of field, muted neutral office background out of focus, natural skin texture, no smoothing, candid half-smile.

Same subject as attached, unchanged. Outdoor portrait in overcast light, 105mm lens, blurred urban background, natural fabric texture on clothing, visible pores and fine lines, warm but realistic skin tone, no plastic sheen.

Same subject as attached. Match this reference headshot exactly for light direction, backdrop colour [hex or name], lens and crop: [describe the approved reference]. Only change the pose and expression to [describe].

Take the attached AI headshot and reduce skin smoothing to a realistic level, restore visible pores and fine texture, correct the left ear which looks warped, and add a subtle natural shadow under the jaw and chin.

Questions people ask

Why does the skin always look like plastic?

Most models default to an averaged, smoothed skin texture because it scores well on generic beauty training data. Explicitly asking for pores, texture and no retouching pushes the result away from that default.

Can I use my own photo as a base?

Yes, and you should. Editing a real photo for light and background keeps facial proportions accurate; generating a face from a text prompt alone is far more likely to drift from the real person.

What ruins a headshot fastest?

Flat frontal light with no shadow, glassy symmetrical eyes, and warped ears or glasses. Check those three areas first on every result before judging anything else.