Can AI make a portrait? Yes: tools, prompts and the rules in 2026

Yes, AI can make a portrait, and in 2026 it's good enough that most people can't tell a careful result from a studio photo. You either upload a photo of yourself and describe the change you want, or you type a description of a person who doesn't exist. Free assistants like ChatGPT and Gemini handle both; paid headshot services like Aragon train on several selfies and return a batch.

The hard parts aren't technical anymore. They're consent (whose face is it?) and copyright (who owns the picture?). I'll cover both after the tools and prompts.

Can AI make a portrait? The short answer

An AI portrait generator takes a reference photo, a text prompt or both, and returns a new image of a face in the style you asked for: corporate headshot, oil painting, anime, Renaissance pet. Quality is high. Likeness is usually close but not perfect, and it gets worse the further you push the style.

For a single picture of yourself, a free chatbot is probably enough. For 40 or more matching headshots, a paid service makes more sense.

Two ways AI builds a face

Under the hood, there are two routes, and knowing which one a tool uses tells you what to expect from it.

The first is reference editing. You upload one photo, and a multimodal model (one that reads images and text together) redraws it to match your instructions while trying to keep your features. That's what happens when you drop a selfie into ChatGPT or Gemini. OpenAI says its Images 2.5 update, launched on 8 September 2026, "more reliably preserves recognizable subjects and distinctive features from reference photos", and that people now make more than 3 billion images a week across ChatGPT and its image API. On Google's side, the Nano Banana 2.1 model card lists fusion of up to 14 reference images and character consistency for up to 4 people.

The second route is personalisation by fine-tuning. Here the service takes a handful of your photos and briefly retrains a model so it learns your face as a new concept. The method most people point to is DreamBooth, a 2022 Google Research paper that showed "just a few images of a subject" were enough to bind that subject to a unique identifier and then render it "in diverse scenes, poses, views and lighting conditions". Aragon doesn't publish its method, but asking for "at least 6 selfies from different angles" and taking minutes rather than seconds fits this pattern.

Text-only portraits are a third, simpler case. With no reference photo, the model invents a face from your words. Handy for book covers, game characters and mood boards; useless if you need it to look like you.

Picking an AI portrait generator

Match the tool to the job. This table covers the main options as of late October 2026, based on the vendors' own pages and published reviews (nobody here ran hands-on tests).

A few notes on the rows. ChatGPT's free tier includes image generation, but Engadget points out that OpenAI doesn't publish the exact limit; there's one weekly allowance you can check under Settings. In Gemini, TechSpot reports that Nano Banana 2 is the default generator across every image mode, and every image carries Google's invisible SynthID watermark. Paid Google plans currently run $4.99 for AI Plus and $19.99 for AI Pro, going by ProPakistani's October price table.

For deeper comparisons, we've ranked the best free AI headshot generators, the best AI photo to painting tools and the best AI pet portrait generators. There's also a full Aragon AI review if you're weighing the paid route.

Real examples: prompts you can adapt

Prompts work best when they say what to keep, what to change, and what the light and framing should be. These are examples to adapt, not magic formulas; swap in your own details.

A LinkedIn headshot from a selfie

Example prompt to adapt: Using the attached photo, create a professional headshot of the same person. Keep my face, hair and skin tone exactly as they are. Navy blazer over a white shirt, plain light grey studio background, soft window light from the left, chest-up framing, eye level, natural expression.

The "keep my face exactly as it is" line matters more than it looks. Without it, models tend to smooth skin and tweak features toward an average face.

A painted portrait

Example prompt to adapt: Turn this photo into an oil painting in the style of a 19th-century portrait. Visible brushstrokes, warm brown background, Rembrandt-style lighting from above left. Keep the person's likeness and expression.

I'd stick to historical styles like this rather than naming a living artist. It's cleaner ethically, and some tools refuse living-artist prompts anyway.

A character who doesn't exist

Example prompt to adapt: Portrait of a woman in her sixties, a retired ship's engineer, short grey hair, weathered face, wearing a wool jumper, standing on a harbour at dusk. 85mm lens look, shallow depth of field, muted colours.

Your pet as royalty

Example prompt to adapt: Using the attached photo of my dog, paint a Renaissance royal portrait. Red velvet cape with gold trim, small crown, dark green background, gilded frame. Keep the markings on her face exactly as in the photo.

Need more help phrasing things? Our guide to writing AI photo prompts goes step by step.

A worked example: the cost of 40 headshots

Say you need a set of consistent headshots for a website, a LinkedIn profile and a conference badge. Here's how the numbers stack up.

Aragon's Basic pack is $35 for 40 photos while its 20% sale runs, which Atlas Cloud worked out to $0.88 per photo; Standard ($45 for 60) and Executive ($75 for 100) both come to $0.75. List prices are $44, $56 and $94. Aragon itself claims on its plans page that a one-hour US photoshoot costs $250 on average, so even at list price you're paying under a fifth of that (though you'd want to check that $250 figure against local photographers before trusting it).

The ChatGPT route costs $0 to $8 a month. But you get one image at a time, and you're the art director. If each usable headshot takes five prompts and you want 10 keepers, that's around 50 generations, which could easily hit an unpublished weekly cap on the free tier.

My rough rule: under five final images, use a chatbot. Over 20 matching ones, a pack service is probably worth $35. In between, it depends how fussy you are.

Benefits and limitations

The benefits are obvious enough. It's cheap, it's fast, and you don't need to leave the house or own a good camera.

The limits are less obvious, so here's where it breaks.

• Likeness drift: big style jumps (anime, heavy painting) pull faces toward a generic look, and small things like eye shape or a gap in your teeth go first.
• Group shots: Google's newest model card caps character consistency at 4 people, so a family portrait of six is asking for trouble.
• Hands, glasses and jewellery: they still warp more often than faces do, especially at the edges of the frame.
• Resolution: Aragon's standard packs are 896 x 1088 pixels, fine for LinkedIn but small for a large print; its Executive pack doubles that to 1792 x 2176.
• Batch quality: on Trustpilot, Aragon showed a 4.9 score across 6,467 reviews in August, yet Atlas Cloud noted recurring complaints about part of a batch looking artificial.

There's a privacy cost too. Engadget reports that OpenAI enrols free ChatGPT accounts in model training by default, so turn off "Improve the model for everyone" in Data Controls before you upload your face. And with Aragon, the full refund only applies if you haven't downloaded your photos, according to its pricing FAQ. Read that before you click download.

Consent when you use other people's photos

Your own face is the easy case. Someone else's isn't, and I think this is where most people get careless.

Start with the tools' own rules. OpenAI's usage policies ban the "use of someone's likeness, including their photorealistic image or voice, without their consent in ways that could confuse authenticity". In practice, a birthday painting of your partner made with their blessing is fine; a realistic fake photo of a colleague is not.

The law is catching up quickly. In the US, the FTC started enforcing the TAKE IT DOWN Act on 19 May 2026, and GRC Report explains that covered platforms must now remove nonconsensual intimate images, AI-made ones included, within 48 hours of a valid request. Euronews reports that creating or requesting nonconsensual intimate deepfakes became a criminal offence in England and Wales in February, and that France bans sharing AI images of someone without consent when it isn't obvious or disclosed that they're fake.

So be blunt with yourself: making intimate or deceptive images of a real person without their consent is harmful, and it's illegal in a growing number of places. Ask first. With kids, ask their parents, and think twice even then.

Copyright: who owns an AI portrait

Short version: in the US, probably nobody owns the purely AI-made parts.

The US Copyright Office's January 2025 report concluded that prompts alone don't make you an author, saying "[p]rompts essentially function as instructions that convey unprotectable ideas", as Mayer Brown summarises. On 2 March 2026 the Supreme Court declined to hear Thaler v. Perlmutter, leaving in place the rule that copyright needs a human author. Your own edits, retouching or a hand-painted layer on top can still be protected; the raw output, likely not. The Copyright Office's AI hub collects the full reports.

The input photo is a separate question. Whoever took the original picture usually owns its copyright, so a portrait built from a professional photographer's shot (or a stranger's Instagram post) can step on their rights even if the face is yours. Rules differ outside the US, and if you plan to sell prints, I'd get proper advice rather than rely on an explainer. Our piece on whether you can copyright AI art covers the wider debate.

Practical takeaways for you

A quick checklist before you generate anything:

• Use a sharp, evenly lit, front-facing photo; the output can't be better than the input.
• Pick the route by volume: a chatbot for one image, a pack service for dozens.
• Write what to keep as well as what to change.
• Check training and retention settings before uploading a face.
• Only use other people's photos with their clear permission.
• Don't assume you own the output; keep your own edits if rights matter.
• Label realistic AI portraits when you post them, especially on professional profiles.

That last one is my opinion, not a legal rule everywhere. But a headshot that's noticeably better-looking than you in person can backfire at an in-person interview, for what it's worth.

Next up: stricter likeness laws

Expect stricter likeness rules. Denmark introduced a bill in October 2026 that would ban sharing realistic deepfakes of someone's face or voice without consent, and its culture minister hopes it'll take effect at the start of 2027, Euronews reports. EU countries must also criminalise creating and sharing sexually explicit manipulated material without consent by 14 June 2027 under the bloc's violence against women directive.

On the tools side, the trend is toward better likeness and more references per image. I'd guess the gap between a free chatbot and a paid headshot pack keeps shrinking, which will squeeze pack services on price. Watermarks like SynthID will probably matter more too, as platforms start checking for them.

FAQ

How to create an AI image of yourself?

Open a free assistant like ChatGPT or Gemini, upload a clear, front-facing photo of yourself, and describe the result you want: outfit, background, lighting and style. Tell it to keep your face unchanged. Then refine one detail at a time in the same chat. For a large set of matching headshots, a paid service like Aragon works from six or more selfies.

How to make an AI picture of yourself?

It's the same process: one good photo plus a specific prompt. Pick the style first (headshot, painting, cartoon), then name what should stay the same. If the face drifts, ask for a smaller change or try a less extreme style. Check the app's privacy settings first, because some services use uploads for model training by default.

Is there a free AI portrait generator?

Yes. ChatGPT and Gemini both include image generation and editing on their free tiers, with usage limits that neither company publishes as a fixed number. Paid upgrades start at $8 a month for ChatGPT Go and $4.99 for Google AI Plus.

Can I make an AI portrait of someone else?

Only with their consent, and never in an intimate or deceptive way. OpenAI's rules ban non-consensual use of someone's photorealistic likeness, and laws in the US, UK and France now cover intimate or undisclosed deepfakes of real people.