Quick Answer

Yes, you can. First get the person's written consent to an AI likeness: a separate signed agreement that says what images will be made, where they will run, what the person is paid and for how long. An ordinary model release from a past shoot does not cover new images they never posed for. Then save the person once in a tool that keeps a reusable person. For product photos, Nightjar, Photoroom, FASHN, WearView and Flair save a real person from one or a few photos. HeyGen and Synthesia make talking video presenters, and OpenArt, Replicate and fal keep a character or train a custom image model. Midjourney and ChatGPT keep no saved person, so the person's photos go with each new request. If the images reach people in the EU, label them as AI-generated.

Which platforms can create a reusable AI model of a real person

Four kinds of tools can reuse a real person's likeness, and they make different things. Pick by what you need to make.

• Product photo tools that save the person. You upload a few photos once and pick the saved person for each new image of your products. Nightjar, Photoroom, FASHN, On-Model, WearView, Uwear, Claid, Flair and Botika all offer this in some form. This is the route for catalog, marketplace and ad images, where the product has to stay exactly as it is.
• Video twins. Synthesia, HeyGen, D-ID and Tavus build a talking presenter from a few minutes of video. Synthesia and HeyGen make the person record a consent statement on camera for a video twin. HeyGen's own developer docs say its photo avatars get no such check: "HeyGen runs no consent check and keeps no consent record for a photo avatar". These tools mainly make presenter videos, though HeyGen's Product Placement also makes still images from one face photo and one product photo. For the voice, ElevenLabs bans cloning a living person's voice without their explicit consent (use policy), and its Professional Voice Cloning works only on your own voice: "Even with their consent, you cannot clone someone else's voice" (help).
• Saved characters and trained models. OpenArt keeps a character made from one photo, a prompt or presets. A training run (a LoRA, for example) teaches an image model one face from a set of photos: at least 2 and ideally 10 or more for a LoRA on Replicate, 20 to 80 for Higgsfield Soul ID. Some stay inside the tool, like Higgsfield Soul ID; a LoRA trained on Replicate can be downloaded and used elsewhere. Adobe Firefly custom models list "Photoshoot of a person" as Enterprise only. Leonardo trains an Element on 10 or more photos, though it now files Element training under legacy features. Civitai, often suggested for a LoRA of a person, no longer allows one: its terms prohibit the likeness of real people, "private individuals" included, "in any context".
• General image tools with reference photos. ChatGPT and Midjourney keep no saved person: you attach the person's photos to each new request. Many guides still describe Midjourney's Character Reference, but its Edit Model now replaces it and takes up to 4 reference images per prompt. Gemini works the same way, though in the US its app can also use people you have named in your own Google Photos. OpenAI's service terms say "You may not use Visual Capabilities to reproduce the likeness of any person without express consent and all necessary rights."

Whichever tool you use, getting the person's agreement is your job, and in Nightjar a custom Fashion Model of a real person needs the right to use their likeness. A video twin's consent clip shows the person agreed to the avatar, but it doesn't set out the scope, pay and duration that New York, for one, requires in writing. That is why the written agreement described below matters whichever tool you pick.

Product photo tools that save a real person, compared

For catalog, marketplace and ad images, a product photo tool that saves the person is built for the job: you pick the saved person for each new product image instead of attaching their photos again. Here is how each tool saves a real person, from its own pages in 2026.

HeyGen, Synthesia, OpenArt, Midjourney and LoRA tools compared

Outside product photography, these are the tools most often suggested for this question. They differ in what they make, whether they save the person and how they treat your uploads. From their own pages in 2026:

What this means for product photos. Midjourney and ChatGPT keep no saved person, so each new chat or request needs the person's photos, the product photos and the description again. A model trained on the person stays tied to the one image model it was trained on, wherever you run it. Every image model is best at some shots and weaker at others. Nightjar has a sophisticated routing system that always picks the best AI image model for the task at hand. Each shot goes to the model that does that kind of shot best, which gives better results than any one model used on its own. In Nightjar the product, the look, the person and the setting are saved and carry over to the next product. Nightjar is built to keep your product true to the original: it automatically looks at every generated image and retries the ones that misrepresent the product, at no cost to you. The other tools fit other jobs: HeyGen and Synthesia when the person needs to speak on camera, a trained LoRA when your team runs its own image pipeline, and Firefly when an Enterprise team already works in Adobe apps.

How to choose a tool for a real person's likeness

Choose by what you make, then check how the tool treats the person's photos, because those are what your agreement promises to protect.

• Stills or video. Product pages, marketplaces and ads need still images with your real product. A talking presenter needs a video twin.
• Training on uploads. Read the terms and the privacy policy, not only the feature page. Photoroom uses app uploads to train its models unless you switch that off in your account settings, HeyGen may unless you ask to opt out, and Midjourney's training summary says user inputs were used. FASHN, WearView and Adobe Firefly (for Creative Cloud subscribers) say they don't, and Nightjar does not train or fine-tune AI models on customer uploads or outputs.
• Who owns and sees the images. Flair's terms say only Scale-plan images are private: images on Free, Pro and Pro+ "will show up in Flair Gallery and other Flair users may also use them in personal and commercial settings", and Flair owns Free-plan images. Both are hard to square with a likeness agreement. With Nightjar, you own the images you make.
• Removing the person. When the agreement ends, you need to delete the saved person and take its images out of use, so check how the tool handles both before you start.
• Cost at your volume. Flair includes one custom model on its Free plan, Claid charges nothing extra for your own model, and Uwear charges 1 credit with no subscription and a $17 minimum purchase. Photoroom needs a paid plan, On-Model and WearView need Pro, and FASHN needs Agency, $99 a month, to upload a face in its Studio, though its Agent takes one on any plan. Nightjar includes custom Fashion Models on every account.
• A trial on your hardest product. Before you commit, make a few images of the consenting person with the product that is hardest to get right for your brand, and compare them with real photos of the person.

What consent you need before making an AI model of a real person

You need the person's own written agreement to an AI likeness, separate from any photo release. Laws differ by place, but the ones that apply to fashion and advertising point the same way: written and specific, and in New York, paid.

An old model release doesn't cover it. A typical release covers the photos taken at a shoot. New York's Department of Labor answers this exact case in its Fashion Workers Act FAQ. A model who agreed to AI use for one job must approve a new campaign in writing: "If initial approval did not cover a new campaign, you must obtain new approval from the model."

New York. Since June 19, 2025, the Fashion Workers Act requires a client to "obtain clear and conspicuous prior written consent from the model for any creation or use of a model's digital replica" (Labor Law § 1037). The consent must detail "the scope, purpose, rate of pay, and duration of such use." The state can fine a client $3,000 for a first violation and $5,000 after that (§ 1038). The Act covers professional models; for anyone else in an ad, including staff and founders, Civil Rights Law § 50 already requires written consent to use a living person's likeness for advertising or trade.

California. Civil Code § 3344 requires prior consent to use someone's likeness to sell products, with damages of $750 or the actual harm, whichever is greater. Labor Code § 927 makes a digital replica clause unenforceable when it lacks "a reasonably specific description of the intended uses" and the person had no lawyer or union negotiating it. SB 1111, signed in 2026 and expected to take effect on January 1, 2027, brings digital replicas into § 3344 and removes a presumption that shielded incidental use of an employee's likeness (Governor's release, Senate analysis).

Other US states. Illinois requires "previous written consent" for commercial use of a person's photograph or likeness (Right of Publicity Act). Tennessee's ELVIS Act makes a person's name, photograph, voice and likeness their property (HA0578). There is no federal digital replica law yet: the NO FAKES Act of 2026 (S.4591) passed the Senate Judiciary Committee and has not become law.

EU and UK. Photos of an identifiable person are personal data under GDPR Article 4, and so are AI images that show them. You need a lawful basis, usually the person's consent or a contract with them. Consent can be withdrawn "at any time", and "It shall be as easy to withdraw as to give consent" (Article 7(3)). For staff, the European Data Protection Board says consent is "unlikely to be freely given" (guidelines), so taking part has to be truly optional. The UK has no general image right, as the Court of Appeal held in Fenty v Arcadia, so the contract and UK GDPR do the work there. Denmark introduced a bill in 2026 that would protect a person's face and voice from realistic digital imitations (Ministry of Culture); it is not law yet.

Germany and France. Germany requires a person's consent to publish their picture (KUG § 22), and its Federal Labour Court has held that employees must give that consent in writing, separate from their job (8 AZR 1010/13). In France, the right to one's image comes from Article 9 of the Civil Code, and the Cour de cassation upheld a model's release because it clearly limited its duration, territory, media and contexts. French criminal law also punishes publishing an AI-generated image of a person without their consent unless it is obviously AI-made or says so (Article 226-8). Paid influencers in France must also mark AI images of a face or body "Images virtuelles".

Lookalikes. Using someone because they look like a celebrity can still trigger publicity claims. Our answer on AI models that resemble a celebrity or influencer explains the risk.

This is general information, not legal advice. A lawyer who knows advertising and employment law where you sell should review the agreement.

What to put in an AI likeness agreement

Write a separate, signed agreement for the AI likeness rather than a clause inside a release or an employment contract. These terms come from what the laws above ask for, and from what tends to go wrong later.

Our help answer on likeness fees for an AI fashion model covers how pay is usually set.

Do AI images of a real person need an AI label

In the EU, yes, for photorealistic images. Since August 2, 2026, the EU AI Act requires anyone who publishes a "deep fake" to disclose that it is AI-generated (Article 50(4)). The European Commission's guidelines on Article 50 count "digital replicas of real persons" as persons for this rule. Article 50(4) has no exception for consent, and the guidelines add that showing a real person still has to respect their rights over their image. The label has to be visible when someone first sees the image; metadata hidden in the file does not count (Commission Q&A). A company outside the EU whose ads are shown in the EU is covered too. Fines go up to €15 million or 3% of worldwide turnover.

The same guidelines say a real product photographed against an AI-made background is generally not a deep fake, as long as it doesn't mislead. The label is about the person.

New York's synthetic performer law, in force since June 9, 2026, does not cover your real model. It requires a disclosure in ads for a "synthetic performer", which it defines as a performer "who is not recognizable as any identifiable natural performer" (General Business Law § 396-b). An AI image of your recognizable brand model is governed by consent law instead. Disclosing anyway costs little.

In 2026 Amazon began asking sellers to tag listing and A+ images that contain "photorealistic AI-generated people" with metadata, and it adds an AI indicator to tagged images where it applies. Its announcement exempts images that "only feature real people (even if they have been altered with AI tools)". It doesn't say where a new AI image of a real, consenting model falls, so ask Seller Support before you skip the tag. To tag an image, add the keyword contains-synthetic-performer to the file's XMP dc:subject field with a metadata editor before upload; our answer on Amazon's rules for AI product images covers the rest.

How to create a reusable AI model of a real person in Nightjar

A custom Fashion Model is a saved person your whole Team can pick for any product. Nightjar makes it from 1 to 5 photos, with no training run.

1. Sign the agreement first. Get the person's written consent to an AI likeness, with the terms above.
2. Choose the photos. Use 1 to 5 recent photos of the same person, face clearly visible, in soft even light, without sunglasses or heavy filters. Include their hair as you want it to appear.
3. Save the person. Open Models in the sidebar (Studio > Models on phones) and choose New model. Enter a name, an Age Range and a Gender, upload the photos and choose Create model. Creating and reusing a Fashion Model shows each screen.
4. Make the first images. In Create, add a Product, choose With model and pick the Fashion Model. A Fashion Model sets the person's face and hair. Put build, height and styling in Custom Directions, such as "slim build, 170 cm, relaxed tailoring". With Add Fashion Model, up to three saved people can appear together, such as your founder and your house model.
5. Save the setup as a Recipe, so the next product starts from the same person, pose, setting and styling. Saving and applying a Recipe walks through it, and reusing one Fashion Model across a collection covers whole collections.
6. Review before publishing. Nightjar automatically looks at every generated image and retries the bad ones at no cost to you. Compare the first images with real photos of the person, and send them for approval if your agreement gives the person that right.
7. Label and log. Add a visible AI label where the law asks for one, and keep a record of which images used the likeness.

When the agreement ends or the person withdraws, open Models, choose Delete on the Fashion Model's card and confirm. That removes it from future use by the Team, and it can't be restored. The source photos and the images already made stay in your Library as separate files, so take those out of use as the agreement requires. Our answer on training an AI model to look like your real model covers editing a custom Fashion Model too.

Our legal guide to AI product photography covers copyright and platform rules, and controlling AI fashion models covers age, body type and pose. To try a custom Fashion Model with someone who has agreed, start with one product photo. New accounts get 6 free Credits with no card.