What you’ll be able to do

Right, so you need a professional photo. LinkedIn’s asking for one, or your CV, or your company’s “meet the team” page, and you don’t have a photographer, you don’t have a studio, and you definitely don’t have the budget for either. Fine. This is doable with your own phone and an AI tool or two, some free, some paid once, and I’m going to walk you through it properly.

The one thing I need you to understand before we start: the goal is not the most flattering photo AI can invent. It’s a photo that a colleague would look at and go, yep, that’s them. That distinction matters more than anything else in this guide.

By the end of this you’ll have a professional headshot built from your own photos, one that still looks like you, ready to upload to LinkedIn, attach to a CV, or use on a work profile. Not a generic AI face wearing your haircut. You.

Before you start

There are three routes here, and they trade off against each other in a pretty simple way: the more editing you hand over, the more of your actual face is at risk.

Route one is retouching a real photo. You keep the photo you already have, you just change the background, tidy the lighting, maybe swap what you’re wearing. Nothing about your face changes. This is the safest route by a mile.

Route two is asking a general AI assistant, ChatGPT, Gemini, that sort of thing, to edit a photo you upload. You’re trusting the model to keep your face intact while it changes other things. Usually fine. Sometimes it drifts, and I’ll show you how to catch that.

Route three is a dedicated headshot service, like Aragon or HeadshotPro. You upload a few selfies and it generates dozens of brand new photos in different outfits and settings. That gives you the most polished-looking set, but every one of them is a new picture made from your selfies, not an edit of a photo you took. That’s a bigger leap, and it’s where the “that’s not quite me” problem shows up.

None of these is “wrong”. Pick based on how much you trust the tool, how much time you’ve got, and honestly, how nervous you are about your face being fed into something you don’t fully understand. If you’re nervous, start with route one. You can always go further later.

Get set up

Before any of this works, you need decent source photos. Doesn’t need to be complicated.

Find natural light, ideally near a window, and face it rather than having it behind you. No harsh shadows across your face. No heavy filters, obviously, the AI needs your actual features, not a smoothed-out version of them. Plain background if you can manage it, a wall’s fine. Simple clothes, nothing with a busy pattern. Phone at eye level (not below, we’ve all seen those photos), and frame it from the shoulders up.

If you’re going the dedicated-service route, take several: different angles, different expressions, maybe one smiling, one neutral. The number each service wants varies, so follow what its upload screen asks for rather than guessing.

Now, the tools. For route one, Adobe Photoshop’s Generative Fill works well if you have access to it: you select an area, type what you want changed (or leave the prompt blank and let it fill from the surrounding pixels), pick a variation, and the edit is non-destructive, so your original stays underneath. For route two, ChatGPT (on any tier, web or the app) and Gemini can both edit an uploaded photo when you describe the change. For route three, Aragon and HeadshotPro are the two I’d look at first.

One thing before you upload anything anywhere: your face is personal data. Aragon’s own privacy policy says face data may count as biometric data depending on where you live, and it asks for your consent before it processes it. Take that seriously whichever tool you use. So a few habits worth building now. Only ever upload your own face, never someone else’s, even as a joke. Read the tool’s deletion policy before you upload, not after. In ChatGPT, turn off “Improve the model for everyone” in your data controls so new conversations aren’t used for training. In Gemini, check your Keep Activity setting: while it’s on, some chats are read by human reviewers, and Google keeps reviewed chats for up to three years even if you delete your activity. And never, under any circumstances, upload an ID document or passport photo to any of these. More on why in a minute.

Try it yourself

Start with the lightest touch, since it’s the safest. Say you’ve got a decent photo already but the background’s a mess and the lighting’s flat. Upload it to ChatGPT or Gemini and try something like this:

This is a photo of me. Keep my face, features, skin tone, hair and expression exactly as they are. Only change the background to a plain soft grey studio backdrop and even out the lighting on my face. Do not smooth my skin or change my face shape.

That last sentence is doing a lot of work. Without it, some models will quietly “improve” your skin, which is exactly what you don’t want.

In ChatGPT you can also highlight just the background with the selection tool before you describe the change. OpenAI’s own help page warns that highlights aren’t always precise and edits can spread beyond the area you picked, so check your hairline afterwards.

If you’re using Photoshop’s Generative Fill instead, you’d select just the background with a selection tool like the Selection Brush, keep your face outside the selection, and type a similar prompt. The fill goes into the area you selected, which is why this is the route where your face is least likely to change.

Route two, asking a general assistant to build something closer to a full headshot from a few of your photos, needs a bit more detail in the prompt:

Here are three photos of me. Using the first as the main reference, make a head-and-shoulders professional headshot: navy blazer, white shirt, plain light background, soft window light from the left. It must look like me, not an idealised version. If you cannot keep my face the same, tell me rather than guessing.

Gemini specifically lists something called character consistency, which is meant to keep a person’s look stable across generated images, so it’s worth trying there. Worth knowing too that Gemini’s image editing is restricted to users 18 and over, and the personal Avatar feature isn’t available if you’re in the UK, EEA or Switzerland, so don’t go looking for that one.

Route three is more hands-off. You sign up to Aragon or HeadshotPro, upload your selfies where it asks, pick some outfit and background options, and wait. Aragon generates up to 100 headshots in roughly 15 to 45 minutes depending on the package, priced as one-off packages in US dollars in the tens of dollars; check aragon.ai/pricing for current numbers. Its privacy policy says it deletes your face data within 30 days of delivering your photos (it can be kept longer if you’re on a subscription, and never more than a year), and you can delete it yourself any time from your account settings. HeadshotPro gives you 30 to 70 headshots depending on package, also a one-off payment in the tens of dollars, check headshotpro.com/pricing. It says outright it never trains its AI on your photos, and both your uploaded selfies and the generated results auto-delete after 30 days, with download links lasting 7 days, so save what you actually want to keep.

Check the result

Right, this bit actually matters more than the generating. Don’t just glance at it and think “yeah, looks alright” and move on.

Show it to someone who knows you, without telling them it’s AI-generated. See what they say before you say anything. Then put it side by side with a real recent photo, same size on screen, and go through it properly: eyes, nose, teeth, hairline, skin tone, your actual age (AI edits can smooth years off you, which sounds nice until a colleague does a double take in person). Check for the classic tells too: melted or asymmetric earrings, a collar that doesn’t sit right, teeth that look slightly too even or slightly too many.

Ask the assistant to help you spot the differences directly:

Compare this edited headshot with the original photo of me. List every difference in my face: eyes, nose, mouth, teeth, jawline, hairline, skin texture, age. Be blunt; I need it to still look like me.

And here’s the actual test that matters: pick the version that looks like you on a normal day, not the one that looks most flattering. Flattering isn’t the goal. Recognisable is.

If it doesn’t work

Face drifting after an edit is the problem you’re most likely to hit. Go back to the original photo, not the edited one, and ask for a single change at a time rather than several at once. State plainly what has to stay the same. If it keeps drifting, drop back to route one and use a tool like Photoshop’s Generative Fill where your face is physically outside the selection.

Something like this works well for pulling it back:

That changed my face. Go back to the original photo I uploaded and make only one change: replace the background with a plain off-white wall. Leave everything else untouched.

If a dedicated service gives you results that look vaguely like a cousin of yours rather than you, the fix is usually better input photos: more angles, more even lighting, no filters at all. Failing that, use the service’s own redo or refund process rather than settling.

If a background edit bleeds into your hair, make the selection smaller and tighter, or switch to Photoshop where you control the selection edge directly rather than describing it in words.

And on privacy, a proper warning: never hand any of these tools an ID document, a passport photo, or anything meant for official use. UK passport photos must be unaltered by computer software, so an AI-edited or retouched headshot doesn’t meet the rules and your application gets delayed. Get a passport photo from a booth or shop, or take a plain one during the application, and leave AI out of it. Same thinking for any other ID.

On LinkedIn, the Professional Community Policies say not to use an image of someone else, or any image that is not your likeness, as your profile photo. So an AI headshot is fine there, and on a CV or an internal staff page, as long as it passes the colleague test above. If an employer, a client or a platform asks for an unedited photo, give them one. And if someone asks whether your headshot is AI, just say so. It’s a photo of you made with a tool, nothing to hide.

Questions people ask

Can I make an AI headshot for free?

Partly. ChatGPT's image editing works on any tier, web or app, and Gemini can edit an uploaded photo too. Dedicated services like Aragon and HeadshotPro are one-off payments in the tens of US dollars, so check their pricing pages for current numbers.

Which AI tool is best for a professional headshot?

For retouching a real photo, Photoshop's Generative Fill is safest, because you keep your face outside the selection. For editing an uploaded photo by description, use ChatGPT or Gemini. For a full generated set, Aragon or HeadshotPro are the two to look at first.

How do I stop AI changing my face in a headshot?

Tell it plainly: keep my face, features, skin tone, hair and expression exactly as they are, and do not smooth my skin or change my face shape. If it still drifts, go back to the original photo and ask for one change at a time.

How do I check if my AI headshot still looks like me?

Show it to someone who knows you without saying it's AI-generated, then put it side by side with a recent real photo and check eyes, nose, teeth, hairline, skin tone and age. Pick the version that looks like you on a normal day, not the most flattering one.

Can I use an AI headshot for a passport photo or LinkedIn?

Not for a passport: UK passport photos must be unaltered by computer software, so an AI-edited one doesn't meet the rules and delays your application. LinkedIn is fine, as long as the image is your own likeness and passes the colleague test.

Keep exploring

If you want to take the background or lighting edit further, there’s a full guide on “How to get AI to edit a photo”. Once your headshot’s sorted, the natural next step is “How to get AI to update my LinkedIn profile”. And if the CV’s still a mess, there’s “How to get AI to write my CV” as well.

Sources and review notes

Review date: 26 September 2026 — every source above was opened and checked against the text on that date.