Linvi← All articles

September 20, 2026 · 9 min read

Best Photos to Upload for AI Headshots (2026)

The best photos for AI headshots are decided before you generate anything. A practical shot list — angles, lighting, variety — so results still look like you.

By Alexander Willard, Founder of Linvi

Here's the thing nobody tells you: your AI headshots are already good or bad before you click generate. The model can only work with what you feed it. Garbage selfies in, uncanny wax figure out.

So the whole game — the entire game — is picking the best photos for AI headshots up front. Not the tool. Not the prompt. The source images. Get that right and even a modest generator returns something you'd happily put on your LinkedIn profile. Get it wrong and no amount of re-rolling will save you.

This is the shot list I'd hand a friend before they upload a single thing. Exact angles, lighting, expressions, and the specific mistakes that produce that plasticky, "that's not quite me" result.

Why your source photos decide everything

An identity-preserved AI headshot system doesn't invent your face. It studies the photos you give it, builds a sense of what you actually look like, then re-lights and re-frames that person in new settings. Think of it as a very literal-minded portrait painter who has only ever seen the reference photos on the desk.

If every reference photo is a dim bathroom selfie shot from below, the painter concludes that is your face. Low light, nostril angle, and all. Then it dutifully reproduces those flaws in a nice blazer.

So the quality ceiling gets set by your uploads. This is also why AI headshots and a studio session aren't as different as people assume — both live or die on lighting and angles. I broke down where each one wins in AI headshots vs. a photographer, but the short version: a photographer controls the light in the room, and with AI you control it by choosing better source frames.

The 10–15 photo shot list

Most good systems want somewhere between 8 and 20 photos. More isn't automatically better. Fifteen varied photos beat thirty near-identical ones, because variety is what teaches the model what stays constant about your face across different conditions. That's the whole trick to likeness accuracy.

Here's what to include.

Angles: cover the full range

Avoid anything shot from way below (the up-the-nose selfie) or way above (the classic dating-app hover). Extreme angles distort your proportions, and the model treats that distortion as your actual bone structure.

Lighting: soft, even, and from the front

Good light is boring light. You want your face evenly lit with no harsh shadow cutting across it.

What to skip: hard midday sun (raccoon eye shadows), a single lamp off to one side (half your face in the dark), and colored light. That purple gaming LED strip will bleed onto your skin and the AI will keep it.

Expression and variety

Don't upload fifteen photos of the same locked smile. Give the model range:

Mix it up beyond the face too. Different days, different shirts, different rooms. A few recent haircut/beard states if they're all current. What you're doing is teaching the system, "this is me across many moments" — that's what produces natural-looking identity-preserved AI headshots instead of a single frozen mask.

Framing and resolution

What NOT to upload (the waxy-look list)

This section matters more than the last one, honestly. Most bad AI headshots trace back to a few specific bad inputs.

On the waxy problem specifically: it's not usually the AI being "too smooth" by nature. It's that you fed it smooth. Retouched selfies teach it that your skin has no texture, so it faithfully returns a mannequin. Feed it honest, slightly imperfect photos and the output looks like a photo, not a render.

If you're worried about the fake tell — the thing recruiters clock in half a second — I went deep on the specific giveaways in do AI headshots look fake to recruiters. Skin texture is the number one culprit, and it starts at the upload stage.

A worked example: from bad batch to good batch

Let me make this concrete. Say you're a founder who needs a headshot for a fundraising deck by Friday.

The lazy batch (don't do this): You scroll your camera roll and grab whatever's there. Three shots from the same wedding. Two group photos cropped to just you. One flattering filtered selfie. A dark bar photo where you look great to human eyes but the AI sees mud. Result: outputs that look 70% like you, weirdly smooth skin, and a jaw that's not quite yours. You re-roll ten times. Still off.

The 20-minute batch (do this): You stand by the living room window at 2pm. Phone at eye level, propped on a stack of books. You take:

No filter. Original resolution. Head and shoulders. Sixteen photos, twenty minutes, zero dollars.

The difference in output is not subtle. The second batch produces headshots where your friends say "nice photo" instead of squinting and going "...is that AI?" That's the entire ballgame.

Two quick pre-shoot decisions

Before you shoot your batch, decide two things — because they carry through to the result.

Outfit. Wear what you'd wear to the meeting you're dressing for. The system reads your clothing and can keep or swap it, but a solid, uncluttered top gives cleaner results than a busy pattern. There's a full breakdown of how outfit, background, and expression work together in this guide to profile photo choices.

Background. For your source photos, a plain wall is ideal — it's easier for the model to separate you from the background. You'll pick the final background later, and if you're unsure which one reads as professional, the 2026 headshot background color guide covers what actually works versus what looks dated.

How many photos is enough?

A practical rule: 12 to 16 varied photos is the reliable sweet spot for most people. Below 8 and the model doesn't have enough to lock your likeness. Past 20, you're mostly adding redundancy unless each new photo shows a genuinely different angle or condition.

Quality and variety over raw count. Ten excellent, varied photos will always beat forty near-duplicates.

FAQ

Can I use old photos if they look more like me?

Use recent ones. Your face changes — weight, hairline, glasses, beard. If you upload a mix of 2019 and 2026 photos, the model averages them and produces someone who looks like neither. Pick a consistent, current version of you and stick to it.

Do selfies work, or do I need a "real" camera?

Modern phone selfies are completely fine — often better than a fancy camera used badly. The front camera's slightly wider lens can distort your face at close range, so hold the phone a bit farther away, or use the rear camera with a timer. What matters far more than the camera is lighting and angle variety.

Why do my AI headshots look plasticky and smooth?

Almost always because your source photos were filtered or heavily smoothed. The model learns your skin has no texture and reproduces that. Upload unfiltered photos with visible, natural skin detail. If you want to avoid the waxy AI headshot look, this single fix does most of the work.

Will it change my face to look better than reality?

With an identity-preserved approach, the goal is you, well-lit and well-framed — not a different, prettier person. That's the point. A headshot that looks like a stranger fails the moment you show up to the meeting. Good source photos plus a system built to preserve likeness keeps the result unmistakably you.

How long does the whole thing take?

Shooting a solid batch takes about 20 minutes. Generating and picking your favorites takes another sitting. Budget an hour end to end. The 20 minutes on the front end is where most of the quality comes from, so don't rush it.

Once your photo batch is dialed in, the last mile is turning those images into a headshot — plus the banner, headline, and About section that surround it — without hiring three separate people. Linvi does exactly that in one sitting: professional headshots built from your own photos so the likeness stays yours, alongside a full brand package for a one-time $99. Shoot your 16 photos by the window first, then let it do the rest.