Ads with a made-up model: the rules, the costs and safer options

A model day means a photographer, a model, a studio, hair and make-up, a release form and usage fees for how long and where the pictures run. A made-up face looks like a shortcut around all of that. It can be. But customers and ad platforms may care, and one wrong sentence in the ad script can turn a cheap shortcut into a complaint or a banned account.

This page covers what counts as a made-up model, the one rule that matters most, what the US Federal Trade Commission (FTC, the consumer-protection regulator), Meta (the company behind Facebook and Instagram) and Amazon say, a worked cost comparison, a test you can run and the safer options. Where a real person or a no-AI approach is better, I say so.

This page may contain affiliate links from which Blur Studio earns commissions.

Cheapest and safest first: no synthetic person at all

Before you make a face, ask whether you need one. These options avoid every synthetic-person rule:

• Product-only shots. Clean photos of the item, on a table or in a hand. No face, no rules about faces.
• An illustrated character or mascot. A drawn logo character, or a freelance illustrator for a one-off job. Nobody mistakes a fox for a customer. See how to keep a mascot consistent.
• A friend, staff member or customer with a signed release. A phone photo with written permission can beat a perfect AI face because it is real.
• A freelance model or a UGC creator. UGC means user-generated content, the phone-filmed style of an ordinary customer. Ask for a quote and get usage rights in writing.

If your ad needs a real before and after, a fit check, skin results or sizing, use a real person. A made-up face cannot honestly show those.

What counts as a made-up model, and what is something else

• Made-up model or synthetic spokesperson: a realistic person created by AI who does not exist. This page.
• Virtual try-on: putting your clothes on a model image, real or generated. That is about fit and fabric, a different job from a made-up face.
• AI influencer: a made-up character built to post regularly. See how Higgsfield AI Influencer works, and how to test an influencer idea cheaply first.
• Talking AI actor: a face that speaks your script in video, sold by tools such as Creatify and HeyGen.

The one rule that matters: show or speak for the brand, never claim experience

A made-up person may show your product or say your brand's words. It must never claim a personal experience: "I lost 5 kg", "I have used this for a year", "my skin cleared up". No such person exists, so no such experience happened.

A quick role test: write the ad's first sentence. If the face says "I", "my" or "we tried", rewrite it in brand voice ("Our best seller...", "People ask us...") or use a real customer. Put this on a one-page checklist and check every script against it before you generate anything.

What the FTC, Meta, TikTok and Amazon say

This is information, not legal advice. Rules change, so check the current text and ask a lawyer for anything high stakes.

The US FTC

The Federal Trade Commission's Consumer Reviews and Testimonials Rule took effect on 21 October 2024. In its staff questions and answers, the FTC says:

• There is no blanket ban on AI-generated avatars or virtual influencers.
• An avatar giving a testimonial is a problem when the testimonial is fake or false, and it may still be deceptive under the FTC Act.
• Using a celebrity avatar without permission to speak well of a product breaks the rule if people would think the celebrity said it.
• On disclosures the rule requires, FTC staff say a platform's built-in disclosure tool does not automatically make the disclosure clear and conspicuous. That answer is about insider-relationship disclosures, but the same logic is a good reason not to trust a toggle alone.

The staff guidance is not a safe harbor, and civil penalties can apply to knowing violations. The takeaway matches the one rule above: do not present a synthetic person as a real, named customer.

Meta (Facebook and Instagram)

Meta's help centre says it adds an "AI info" label to ad images created or materially edited with Meta's own generative AI features, and that photorealistic AI humans get a label next to "Sponsored". It also says minor edits and ads made with third-party tools may not get the label, and that ads about social issues, elections or politics must disclose digitally created or altered content, including from third-party AI. The page I read is the UK version and may lag the current rules. Do not read "may not get the label" as "no duty": the FTC honesty duty applies whatever the platform label does.

Amazon

Amazon Seller Central announced that product images and videos with photorealistic AI-generated people must carry the metadata keyword contains-synthetic-performer in the dc:subject field (IPTC/XMP, a standard place to store image tags) before you add them to listings and A+ Content. It is not needed for real people, fictional-work characters, images without people or non-photorealistic people. Sellers on the forum report confusion about partial bodies and the tag disappearing from the live processed image. Practical advice: tag the file before upload, keep an untouched original, and log it in a spreadsheet. Check Amazon's current policy page before you rely on this. Other listing rules are in Amazon and marketplace listing videos.

TikTok and the EU

TikTok's ads policy page on misleading and false content says content that is completely AI-generated, or significantly modified by AI, is allowed if you apply TikTok's AIGC (AI-generated content) label or add a clear disclaimer, caption, watermark or sticker of your own. It says minor edits such as lighting changes or background removal do not count, that undisclosed AI content gets the ad rejected or restricted, and that using a public figure's likeness without permission is not allowed (likeness of private people used without permission is removed). One vendor article says the Ads Manager toggle resets when you duplicate a campaign. I could not confirm that on TikTok's own pages, so check the toggle again after you duplicate. For EU AI transparency duties, check the current EU rules. A simple habit helps everywhere: add "AI label set?" to your launch checklist.

Cost: a model day versus a made-up face

An honest comparison counts rejected images and human review, not only the subscription. Use this formula:

• Real model day = (photographer + model + studio + make-up + usage fees) divided by the usable finished images. Get two real quotes.
• Made-up face = your share of the subscription + (images you need divided by your on-brand rate) x cost per image + about an hour of human review.

Illustrative numbers, made up for the example. Suppose a model day quote is 1,500 and gives 30 usable images, so 50 per image. Suppose you need 30 usable AI images and only 1 in 3 is good enough: that is 90 generations. If a generation costs a fraction of a credit, the credits are small, so the real cost is review time. Add an hour or two for checking hands, jewellery, fabric and the stare. The AI route is often cheaper, but the gap shrinks fast if you need many angles of the same person, real fabric fit, or legal review.

Competitor prices, as printed by the vendors on 2026-11-03: Creatify, a tool for making ads with AI actors, lists Standard at $15 a month or $10 a month billed annually, and Pro at $55 a month or $38 a month billed annually. HeyGen, a tool for talking-avatar videos, lists Free ($0, 3 watermarked videos a month), Creator $29, Pro $49 and Business $149 plus $20 per seat. Check each vendor page for the current price. Talking faces are what those tools are made for.

Where Higgsfield fits and where it does not

Higgsfield is stronger for still images and varied scenes than for a long talking-face video. Its own image models include Soul 2.0, a model with fashion-aware presets, and its Soul ID feature can keep one realistic face across images (it needs 20 or more photos, so for a made-up person you first generate varied portraits, then train). See the AI Influencer guide and the Soul ID guide for the steps. Marketing Studio is its template workspace for ads and UGC-style clips, and you can read about it in the Marketing Studio guide.

Facts to know before you pay, from Higgsfield's pages as of November 2026 (check the current pricing page): a Free account has no credits and its outputs carry a watermark. Starter is $19 a month with 270 credits. Some tools, such as the UGC builder, are marked as included on Plus and Ultra only, so check the plan table. Higgsfield says it does not guarantee outputs are unique, and that invisible machine-readable AI markers may be added to outputs on any plan. Soul ID cannot be exported as a file, so the face stays inside your Higgsfield account. Higgsfield's help center says protected characters, brands and logos can be blocked, and it asks you to upload only photos of yourself or of people who gave permission.

Higgsfield is not the right pick when: you only need a talking face reading a script (Creatify or HeyGen are built for that), you need a real before and after, or the audience is trust-sensitive.

Match the model to your buyers: size, skin tone and fit

For clothing, one body type on every listing leaves many buyers unable to judge fit. WearView, a vendor of AI fashion models, argues that showing a range of sizes and skin tones helps buyers picture themselves in the item and cuts fit surprises. That is a vendor argument and I have not checked its return-rate figures, but the logic is easy to test: show the same garment on two or three body types in your ad variants and watch the comments and returns. A made-up model also cannot prove fit, because the AI draws the fabric, it does not drape the real garment. If fit is your selling point, show the real garment on a real person.

Quality problems to expect

• Hands, rings and jewellery that look wrong.
• Fabric that does not fit or fold the way the real garment does.
• Skin that is too smooth, and a fixed stare.
• The same "type" of face every time, which some customers notice.
• Resemblance to a real person by accident. Run a reverse image search on the final face and show it to three people. If anyone says "that looks like X", regenerate.

Customer trust matters too. Some audiences react badly when they learn a face is AI. Watch the comments for words like "AI", "fake" and "bot".

A 14-day test with pass and fail rules

A plan you can run before committing a budget:

1. Pick one product and write one script in brand voice, with no "I" claims.
2. Make two ads with the same copy and budget: one with the made-up face, one with a real person or a product-only shot.
3. Run both for 14 days. Record cost per result.
4. Pass: the made-up ad's cost per result is within 20 percent of the other ad, and comments show no "fake" or "bot" reactions. Fail: anything else. Switch back to the real or product-only version.

Pre-launch checklist

• Script says nothing a real person would claim from experience.
• Face checked for resemblance to a real person.
• Platform label or toggle set where required, and checked again if you duplicate a campaign.
• Amazon files tagged and an original kept (if selling there).
• No health, finance, beauty-result or children-related claims on a synthetic face.
• A record kept: prompt, tool, date, plan, approvals and who signed off.
• A named human approved the final ad.

Who should skip a synthetic model

• Your claim must be real: results, fit, sizing, safety or health.
• You sell something where trust is the product, such as advice, care or finance.
• You cannot check platform rules for each market you advertise in.
• Nobody can own the approval step and the record file.

How I'd approach this

I would start with the safe options: product-only shots and an illustrated character. If I still wanted a human face, I would run the 14-day test on one product before buying a plan, and I would write every script in brand voice. If the test passes, I would pick the tool by job: a talking face from Creatify or HeyGen, and still-image scenes from Higgsfield. The Higgsfield vs HeyGen guide and the Arcads and Creatify comparison help with that choice, and the credits guide helps with cost. If you want to try it, you can start with Higgsfield here.

If you want the approvals, the record file and the asset folder handled in one repeatable workflow, Blur Studio builds approval portals and shared asset libraries around these tools. You can see what that looks like.