Are AI headshots obvious?
12 November 2026 · 6 min read · PhotoBoost AI
Short answer: in our experience the good ones are not obvious at the sizes headshots are actually used, and the bad ones are extremely obvious.
That is a less satisfying answer than "nobody can tell", which is why most articles selling these tools give you the satisfying one. Here is the more useful breakdown: what makes a generation read as a real photograph, and what gives it away.
Where they hold up
Headshots are used at small sizes. A LinkedIn profile photo is displayed cropped to a circle roughly the size of a thumbnail, and a team page shows a grid of faces at similar scale.
At those sizes, what makes a portrait convincing is lighting, skin texture, eye sharpness and a plausible background — the things modern image models are genuinely good at. PhotoBoost runs on nano-banana, which holds up well on all four.
Our own view is that the output of a good run looks like a photograph at the sizes a headshot is used. We are the vendor, so discount that accordingly — but it is what the output looks like.
Where they break down
The tells are less about uncanny skin and more about specifics:
The asymmetry nobody mentions
A good generation is hard to spot. A bad one is impossible to miss, and one bad photo in a set of twelve is what people remember.
That is why it is worth looking at a full set rather than a single hero image before you pay — the free trial shows you all 12 watermarked photos, which is the honest way to judge it.
People who know you are a different test
A stranger looking at a thumbnail is a much lower bar than a colleague who sees you every day. If someone who knows your face well looks closely and finds something wrong, they will notice.
For most uses — LinkedIn, company pages, CVs, conference bios — the audience is strangers forming a first impression. That is the context where this works well.
Practical checks before you use one
If it fails, try a run with clearer, better-lit selfies — the input quality matters more than anything else.