# Photos of Non-Existent People: How to Create Realistic Portraits Without Image Rights

> How to create realistic photos of non-existent people: what to check to make them believable, what the AI Regulation requires from August 2026, and how to create them with aiPortrait.

- **Source (source):** https://aistohire.com/aiportrait-photorealistic-portraits-ai-60/blog/photos-of-non-existent-people-without-rights
- **Platform (platform):** AIstoHire
- **Language (language):** en
- **Updated (updated):** 2026-11-20

**In summary**

- A "photo of a person who does not exist" is a synthetic portrait: the face does not belong to anyone, so there is no image rights to request or model to pay. What still exists are transparency obligations.
- Random face generators (like *this person does not exist*) give you a face and that's it: no age, no clothes, no scene, no second take of the same person. To work, you need control, not chance.
- From **August 2, 2026**, the transparency chapter of Regulation (EU) 2024/1689 on Artificial Intelligence applies. It affects the labeling of synthetic content and the disclosure of deepfakes. This is general information, **not legal advice**.
- What most betrays an AI-generated portrait are the hands, the eyes, and the "plastic" skin. There are specific decisions in the generator itself that reduce those flaws before pressing the button.
- **aiPortrait**, the mini app for portraits by **AIsToHire**, builds the person with a 4-step assistant (27 outfits, 21 backgrounds, 9 styles, 4 lighting schemes) and leaves 14 editing tools to correct without starting over: [see aiPortrait](/aiportrait-photorealistic-portraits-ai-60).

---

You need a photo of a person. For the mockup of a website, for a character sheet, for a carousel, for the profile of a fictitious client in a presentation. And you always run into the same issue: the stock photo bank charges you, the face you choose is already used by forty other websites, and if you use a real photo you have to ask permission from a real person. This guide explains how to solve it with portraits of people who do not exist: what to control to make them pass, what European regulations say since August 2026, and how to do it step by step with [aiPortrait](/aiportrait-photorealistic-portraits-ai-60), the mini app for photorealistic portraits from AIsToHire.

## What exactly is a "photo of a person who does not exist"?

It is a photorealistic image of a human being generated entirely by an artificial intelligence model, with no real person behind that face. No one posed, no one signed a release, and no one can claim their right to their own image, because that face does not belong to anyone. It is, literally, a face built pixel by pixel from a description.

The concept became popular in February 2019 with **thispersondoesnotexist.com**, the website by Phillip Wang that served random faces of 1024×1024 generated with **StyleGAN**, the generative adversarial network that NVIDIA released in December 2018. Seven years later, the issue is no longer whether the face looks real —it does—, but whether you can **decide what that face looks like**.

And that is where three things that many people confuse separate:

| | Stock photo bank | Random face generator | Custom generation |
|---|---|---|---|
| Does the person exist? | Yes, it is a real model | No | No |
| Can you choose age, ethnicity, clothing? | Only by filtering the catalog | No | Yes, attribute by attribute |
| Can you repeat the same person in another scene? | Only if the model did that session | No | Yes |
| Exclusivity | No: non-exclusive license, the same photo is used by others | Yes (each shot is unique) | Yes |
| Image release / *model release* | Essential for commercial use | Not applicable | Not applicable |
| Cost per image | License or subscription | Free | According to the generator's plan |

The point in the table that is hardest to digest is **exclusivity**. The standard licenses of large stock photo banks are non-exclusive: you pay to use the photo, not to be the only one using it. That is the reason the same smile of the same model appears on your competitor's landing page. With a generated person, the face comes out only once.

## Is it legal to use the face of someone who doesn’t exist on my website or in an advertisement?

Using a synthetic portrait avoids the classic problem —the right to one’s own image of a third party— but it does not leave you in a regulatory void: European AI legislation introduces **transparency** obligations regarding artificially generated content. The following is general and informative information, **it is not legal advice**: for a specific case, consult a professional.

The three regulations worth knowing:

| Regulation | What it says in relation to what matters here | Since when |
|---|---|---|
| **LO 1/1982** (right to one's own image, Spain) | Its art. 7.6 considers it an unlawful intrusion to use the name, voice, or image of a person for advertising or commercial purposes without their consent | May 5, 1982 |
| **GDPR** (Regulation (EU) 2016/679) | Protects the data of identified or identifiable natural persons (art. 4.1); recital 26 excludes information that does not refer to an identifiable person | May 25, 2018 |
| **AI Regulation** (Regulation (EU) 2024/1689) | Art. 50: providers must label synthetic outputs in machine-readable format, and anyone deploying a system that generates deepfakes must publicly disclose that the content is artificial | In effect since 1-8-2024; art. 50 **applies from 2-8-2026** |

Three nuances that make a difference in practice:

1. **If the face does not resemble any real person, the right to one's own image does not come into play.** Art. 7.6 of LO 1/1982 protects the image *of a person*. If the portrait is synthetic and does not evoke an identifiable person, there is no holder who can oppose it. The risk arises when the resemblance to a real person is recognizable.
2. **A synthetic face, in principle, is not personal data.** The GDPR applies to information about an identified or identifiable natural person. An invented face does not identify anyone. It is different when you use parts of **your own photos** or those of a third party to generate the portrait: in that case, you are dealing with data of a real person and need their consent.
3. **Deepfakes have a strict definition.** Art. 3.60 of the AI Regulation defines it as content that *resembles existing real persons, objects, places, or events* and may induce belief that it is authentic. A portrait of someone who does not exist does not obviously fit that definition; the labeling obligation of art. 50.2, aimed at providers of generative systems, is broader. If your use is advertising or informative and the image may lead one to believe it is a real photo of a real person, it is prudent to declare it.

> What **aiPortrait** does **not** do: it does not imprint its own watermarks or write provenance metadata like C2PA in the portraits it generates. If your use case requires declaring that the image is synthetic, you place the notice in the caption, in the credits, or in the file’s metadata.

## A random face won’t do: there are six decisions that determine if the photo passes

The problem with random generators is not the quality of the face, it’s that you cannot request *another photo of the same person*, nor dress them in a medical gown, nor seat them in a café. A portrait that works in a real job is defined by six decisions, all of which are controllable.

1. **Who they are**: gender, age, ethnicity, build, and specific facial features.
2. **What their skin is like**: texture level is, by far, what most separates a credible photo from a render.
3. **What they are wearing**: clothing places the person in a professional context before anything else.
4. **Where they are**: studio background, office, outdoors, or specific scene.
5. **How it is lit**: the lighting scheme completely changes the emotional reading of the same face.
6. **How it is framed**: close-up, medium shot, head angle, and gaze direction.

Point 5 is the one that everyone underestimates the most. This is the same person, with the same framing and the same expression, under the four lighting schemes offered by aiPortrait:

![Comparison in four panels of the same portrait with four lighting schemes: studio, natural window light, golden hour, and high-contrast dramatic light](https://cdn.aistohire.com/i/2028-axzj-ulbw-iphg-txjo.webp)

From left to right: **studio** (diffused softbox, no harsh shadows: the safe option for corporate profiles), **natural** (window light, soft fall: the closest to a spontaneous photo), **golden hour** (warm sun, halo in the hair: sells closeness and lifestyle) and **dramatic** (high contrast, half the face in shadow: editorial, but hides features). Four distinct photos of the same person without touching anything else.

## How to create a photo of a person that doesn't exist, step by step

In aiPortrait the process consists of four steps and ends with a batch of variations, not a single image. You can navigate through them with the selectors or write the description in free text and let the app fill in the assistant for you.

### Step 1 — Define the person

This is the lengthy step and the one that most determines the outcome. The available controls:

| Block | What you choose | Options |
|---|---|---|
| Demographics | Gender, ethnicity, build, height | 3 / 9 / 4 / 3 |
| Age | Exact years with slider | Continuous range |
| Face | Face shape, eyes (color and shape), nose, lips, eyebrows, facial hair | 5 / 6 / 4 / 6 / 3 / 4 / 6 |
| Skin | Tone and texture | 8 tones / 3 textures |
| Hair | Type, color, and hairstyle | 5 / 10 / 9 |
| Marks | Freckles, moles, scars | 4 / 4 |

You can also start from a **reference photo**, a **selfie**, or **3 to 8 photos** to create a reusable character. Important and honest: the platform does not perform fine training of models (neither LoRA nor *fine-tuning*); with those photos, it generates a **canonical portrait** that is then used as an identity reference. The similarity is high, but it is not a clone.

### Step 2 — Dress her

27 outfits spread across five categories: formal (suit, tuxedo, evening gown), business casual (blazer with t-shirt, cardigan, polo), casual (t-shirt, jeans, sweatshirt, linen), sports (gym, running, yoga, soccer) and uniforms from 7 sectors: medical (coat and healthcare uniform), legal, finance, real estate, tech, hospitality (chef) and transport (pilot). Additionally: 11 garment colors plus the option to keep the original color, 8 accessories, and 4 makeup options (including "no makeup").

### Step 3 — Place her in a scene and turn on the light

21 backgrounds grouped into studio (4), office (3), outdoor (4), specific scenes like café, gym, library, restaurant, and home office (5), thematic environments (3), and two special ones: plain white background for cropping and text-customized background. This is complemented by background blur in four levels (from total focus to bokeh f/1.8), light scheme, its direction, intensity, and color temperature.

This is the same person changing only clothes and scene, with the face intact:

![Three panels with the same fictional person changing only clothes and scene: suit in studio, medical coat in clinic, and sweatshirt in a café](https://cdn.aistohire.com/i/9490-qbpb-nlwk-umpd-nbqg.webp)

### Step 4 — Generate multiple shots, not just one

You choose visual style (9: photorealistic, black and white, editorial, corporate LinkedIn, oil, watercolor, illustration, anime, and 3D render), aspect ratio (1:1, 4:5, 2:3, or 16:9), quality, and **number of variations, from 1 to 8**. The app asks the engine for variations that are "distinct frames from the same session": same person and same attributes, with micro-changes in pose, head angle, and light. Working with 4 and discarding 3 is the norm and the healthy approach.

If you're in a hurry, there's a shortcut (**Quick Shot**) that launches a random configuration with a single variation, and five ready-made templates for use cases: professional LinkedIn, blog author, gaming avatar, novel character, and corporate team.

## How to describe the person so that the portrait looks credible

The difference between a weak description and one that works is not in the length, but in how many decisions you leave open. Each attribute that you do not specify is decided by the model for you, and it decides it towards the average: symmetrical face, perfect skin, thirty-something years, catalog smile.

| Weak description | Why it fails | Working description |
|---|---|---|
| "A professional woman" | Everything is open: the average stereotype comes out | "47-year-old woman, Latino origin, oval face, shoulder-length wavy dark brown hair, almond-shaped brown eyes, natural skin with visible pores, navy blue blazer, gray studio, softbox light, 3/4, serious professional expression" |
| "A smiling doctor" | No age, no features, no shot | "58-year-old man, gray-haired, trimmed beard, white coat with stethoscope, blurred clinic hallway, natural light, medium shot, natural smile" |
| "A modern young guy" | "Modern" means nothing to the engine | "22-year-old boy, black afro hair, dark gray sweatshirt, café with warm lights blurred in the background, soft side light, close-up, neutral expression" |

Four rules that come from how the generator is built:

- **Ask for "natural" skin, not "retouched".** Natural texture adds pores, fine lines, and micro-relief; retouched gives an editorial finish that the eye reads as a render. It's the cheapest realism lever you have.
- **Hide the hands if you don't need them.** Choosing *arms crossed* or *hands in pockets* takes them out of frame. Hands are the number one flaw of any image generator, and the most reliable way to avoid weird hands is not to have hands.
- **Increase the background blur.** A high bokeh (f/1.8) not only looks good: it dissolves background details, which is where artifacts and scribbles that look like letters accumulate.
- **If the text and the selectors contradict each other, send the text.** The app interprets your description, fills in the assistant, and alerts you to conflicts instead of resolving them silently.

## What you can fix later without regenerating

Almost all the "almost, but not quite" can be fixed by editing the portrait you already have. aiPortrait brings 14 generative editing operations, and each one **creates a new version**: the original is never lost and undoing is going back to the previous version.

| I want to… | Tool |
|---|---|
| Change the background or clothing | Background · Clothing |
| Change the hairstyle, makeup, or an accessory | Hairstyle · Makeup · Accessory |
| Make it appear older or younger | Age |
| Redo just the face without touching the rest | Face |
| Change the lighting of the already made photo | Re-lighting |
| Smooth or fix the skin | Retouch |
| Remove an object or edit a specific area | Delete · Edit area (painted mask) |
| Expand the frame to another ratio | Expand |
| Make the face unrecognizable | Anonymize |

In addition, there are non-generative adjustments that do not consume generation: brightness, contrast, saturation, hue, warmth, 6 filters (black and white, vintage, warm, cool, film, and faded) and film grain.

![Comparison before and after the same portrait: on the left flat gray background and flat light, on the right blurred office, blue blazer, and warm directional light](https://cdn.aistohire.com/i/2359-kfxw-ggem-gshq-koym.webp)

## What sizes and formats you get

The engine generates in 1024×1024, 1024×1536, or 1536×1024 depending on the chosen ratio. From there, the smart cropping automatically centers the face and gives you the exact size for each site:

| Preset | Pixels | For what |
|---|---|---|
| LinkedIn | 400 × 400 | Profile picture |
| Square | 1080 × 1080 | Instagram post, avatar |
| Vertical Instagram | 1080 × 1350 | Vertical feed |
| Story | 1080 × 1920 | Stories and Reels |
| Widescreen | 1920 × 1080 | Header, cover, video |
| CV | 600 × 800 | Resume, profile |

Export supports PNG, JPG, and WebP, in original size, profile (400 px), 2K (2048 px), or 4K (4096 px), and download in ZIP if you need the whole batch. Two honest precisions: the 2K/4K is a **high-quality upscaling with enhancement** (Lanczos3 algorithm), not a generative super-resolution that invents new detail; and the **transparent background** is obtained by generating on a flat white background and cropping by luminance threshold, so it works very well with clean edges and worse with very loose hair or fine strands.

## The six signs that reveal an AI-generated portrait

Before publishing, review these six points in the order they fail. They are the same ones anyone who knows what to look for checks:

1. **Hands and fingers**: number of fingers, impossible joints, merged fingers.
2. **Eyes**: asymmetry, iris of different sizes, reflections (*catchlights*) that do not match between the two eyes.
3. **Ears and earrings**: one earring on one side and nothing on the other is the most common mistake.
4. **Hair edges**: blurry halo against the background or scissor-cut appearance.
5. **Skin**: if it has no pores or lines, the brain reads it as plastic.
6. **Text in the background**: signs, plaques, and labels appear as scribbles that mimic letters.

aiPortrait injects into **all** its prompts a block of quality rules that specifically address these points: anatomically correct hands with five fingers when visible, symmetrical and sharp eyes with natural reflections, clean hair edges, realistic skin that is never waxy, no text, letters, numbers, logos, or watermarks anywhere in the image, and only one person. That **greatly reduces** the failure rate; it does not eliminate it. Always review before publishing, especially if there are hands in the frame.

## The limits of aiPortrait, clearly stated

Knowing what a tool does not do is as valuable as knowing what it does:

| Limit | Reality |
|---|---|
| Exact clone of a person | No fine-tuning (LoRA/*fine-tuning*). With 3-8 photos it generates a canonical portrait that serves as an identity reference: high resemblance, not identical |
| Exact pose by skeleton | The pose is described with words and options; there is no skeleton control or ControlNet |
| Free clothing color | Closed palette of 12 colors, so the engine respects the requested color |
| Super-resolution | The 2K/4K is Lanczos3 upscaling with enhancement, not generated detail |
| Expanded re-cropping | When expanding, the engine rebuilds the image to the new format; it does not keep the original pixels intact |
| AI content marking | It does not add its own watermark or C2PA metadata |
| Minors | Below 18 years the app enforces a reinforced policy: no makeup, no accessories or adult clothing, and neutral and respectful portrait |

## Frequently Asked Questions

### Can I use photos of non-existent people on my website without paying rights?

Since there is no real person behind the face, there is no image transfer or *model release* to manage. What you should review are the terms of use of the tool you are using and, from August 2, 2026, the transparency obligations of Regulation (EU) 2024/1689 on artificially generated content. This is general information, not legal advice.

### What if the generated face accidentally resembles a real person?

It is unlikely but not impossible, and it is the only scenario where the right to one’s own image comes back into play. If someone on the team recognizes a specific person in the portrait, it is wise to discard that image and generate another one: it costs you a new run and saves you a problem.

### Do I have to disclose that the photo was made with AI?

It depends on the use. The AI Regulation requires disclosure when it involves an ultra-simulation, defined as content that resembles existing real people or events. A portrait of an invented person does not clearly fit that definition, but if the image could lead someone to believe it documents a real person — testimonial advertising, press, customer profiles — it is prudent to declare it. aiPortrait does not provide that notice for you.

### Can I get several different photos of the same invented person?

Yes, and that is precisely what distinguishes a controlled generator from a random face website. You keep the person as a character and reuse them by changing clothes, scene, lighting, and framing; additionally, each generation can return up to 8 variations of the same person in the same session.

### Is it useful for creating a professional profile picture from my selfies?

You can upload a selfie, a reference photo, or between 3 and 8 photos, and the app uses them as identity references to generate portraits with clothing, background, and studio lighting. Keep in mind the real limit: there is no fine-tuning of a model with your face, so the likeness is high but not exact. And if you upload photos of another person, you need their consent.

### How long does it take and how many images come out at once?

Each generation produces between 1 and 8 variations and is processed in queue, so you can continue using the app while they complete; if you refresh the page, the work resumes automatically. It is common to request 4 and keep one.

### Why do hands turn out poorly and how can I avoid it?

Hands are the area where all current image generators fail the most because they combine many small joints in a small space. The practical solution is not to insist on the prompt: it is to choose a framing or pose that keeps them out of shot — close-up, arms crossed, or hands in pockets — and reserve visible hands for when they are essential.

### Can I get the photo with a transparent background to overlay it on another design?

Yes, by exporting in PNG with the transparency option. It works by generating the portrait on a flat white background and cropping by luminance, so the result is very good for clean silhouettes and less refined for loose hair or very fine strands. If you need a perfect cut of a complex hairstyle, be prepared to retouch it.

## What decides the result is not the model: it’s your decisions

Photos of non-existent people have ceased to be a technical curiosity to become a work tool: they solve the cost, exclusivity, and image permission in one go. What separates a usable result from one that exposes the generator is not the model, it’s the decisions — skin texture, light, framing, hands out of shot — and the ability to correct without starting from scratch. **aiPortrait**, in **AIsToHire**, is built exactly around those decisions: you can see it at [aiPortrait — Photorealistic Portraits of People Who Don't Exist](/aiportrait-photorealistic-portraits-ai-60).

> Create your first non-existent person. Describe them, choose clothing, scene, and lighting, and get up to 8 variations with studio-quality photos. Start for free with aiPortrait at AIsToHire.

- [aiPortrait — Photorealistic Portraits of People Who Don't Exist](https://aistohire.com/aiportrait-photorealistic-portraits-ai-60)

## About the platform

**aiPortrait — Photorealistic Portraits of People Who Don't Exist** — an app by AIstoHire. Available at: https://aistohire.com/aiportrait-photorealistic-portraits-ai-60/blog/photos-of-non-existent-people-without-rights

### Other languages

- `es`: https://aistohire.com/es/aiportrait-retratos-fotorrealistas-ia-60/blog/fotos-de-personas-que-no-existen-sin-derechos-de-imagen
- `fr`: https://aistohire.com/fr/aiportrait-portraits-photorealistes-ia-60/blog/photos-de-personnes-qui-nexistent-pas-sans-droits
- `it`: https://aistohire.com/it/aiportrait-ritratti-fotorealistici-ia-60/blog/foto-di-persone-che-non-esistono-senza-diritti
- `pt`: https://aistohire.com/pt/aiportrait-retratos-fotorrealistas-ia-60/blog/fotos-de-pessoas-que-nao-existem-sem-direitos-de-imagem
- `de`: https://aistohire.com/de/aiportrait-fotorealistische-portraits-ki-60/blog/fotos-von-nicht-existierenden-personen-ohne-rechte
- `bg`: https://aistohire.com/bg/aiportrait-sazdavayte-fotorealistichni-portreti-na-hora-s-ii-60/blog/snimki-na-hora-koito-ne-sashtestvuvat-bez-prava-varhu-izobrazhenieto
- `ja`: https://aistohire.com/ja/aiportrait-photorealistic-portraits-60/blog/sonzai-shinai-jinbutsu-no-shashin-shozoken-nashi
- `nl`: https://aistohire.com/nl/aiportrait-maak-fotorealistische-portretten-van-mensen-met-ai-60/blog/fotos-van-mensen-die-niet-bestaan-zonder-portretrechten
- `ko`: https://aistohire.com/ko/aiportrait-ai-portrait-saengseong-60/blog/jonjaehaji-annneun-saram-sajin-chosanggwon-eopsi
- `sv`: https://aistohire.com/sv/aiportrait-skapa-fotorealistiska-portratt-av-personer-med-ai-60/blog/foton-av-personer-som-inte-finns-utan-bildrattigheter
- `da`: https://aistohire.com/da/aiportrait-lav-fotorealistiske-portraetter-af-personer-med-ai-60/blog/fotos-af-personer-der-ikke-findes-uden-billedrettigheder
- `ar`: https://aistohire.com/ar/aiportrait-anshi-sura-shkhsia-waqaia-lashkhas-baldhka-alastnaai-60/blog/suwar-ashkhas-ghayr-mawjudin-bidun-huquq-sura
- `no`: https://aistohire.com/no/aiportrait-lag-fotorealistiske-portretter-av-personer-med-ki-60/blog/bilder-av-personer-som-ikke-finnes-uten-bildrettigheter
- `fi`: https://aistohire.com/fi/aiportrait-luo-valokuvatodellisia-muotokuvia-tekoalylla-60/blog/kuvia-ihmisista-joita-ei-ole-olemassa-ilman-kuvaoikeuksia
- `zh`: https://aistohire.com/zh/ai-realistic-portrait-generator-60/blog/bu-cunzai-de-ren-de-zhaopian-wu-xiangquan
- `pl`: https://aistohire.com/pl/aiportrait-tworz-fotorealistyczne-portrety-osob-z-ai-60/blog/zdjecia-osob-ktore-nie-istnieja-bez-praw-do-wizerunku
- `he`: https://aistohire.com/he/aiportrait-tsro-portrtim-potorialistiim-shl-anshim-am-binh-mlakotit-60/blog/tmunot-shel-anashim-shelo-kayamim-bli-zchuyot
- `cs`: https://aistohire.com/cs/aiportrait-vytvarej-fotorealisticke-portrety-lidi-s-ai-60/blog/fotky-lidi-kteri-neexistuji-bez-prav-k-podobizne
- `id`: https://aistohire.com/id/aiportrait-buat-potret-fotorealistis-orang-dengan-ai-60/blog/foto-orang-yang-tidak-ada-tanpa-hak-atas-gambar
