Being Your Own Brand Ambassador With a Brand Character

Three routes: you, a hired model, or an invented person

If you are the founder, you are the cheapest option. There is no release to chase and no rate to renew, and nobody can withdraw permission later. The cost is permanence: once your face is on 40 listing images, replacing it means regenerating all of them. At the other end, an invented person has no model to pay and no permission that can be withdrawn, but it does not make the file legally clear. No image tool, ours included, certifies that a generated face resembles nobody, and neither image provider indemnifies you.

A hired model costs money twice: the shoot, then the right to build a digital replica. In New York, the Fashion Workers Act has required separate written consent for a model's digital replica since 19 June 2025, covering scope, purpose, rate of pay and duration. It binds model management companies and their fashion-industry clients. Outside that scope, your own release wording decides. Do not assume a standard shoot release carries you. In July 2026 a New York court refused to enjoin a brand whose release permitted images 'composite or distorted in character or form', and did not rule on the merits.

How the character is built

You write a description of the person, at least ten characters, and you may add up to four reference photographs, ten megabytes each. The description is required and the photographs are not, which is how an invented character gets built. Send clear photographs of one adult, face unobstructed, in even light, rather than a collage. Brand characters are adults only, and a character that reads as under 18 is rejected on the server.

One portrait comes back: a single three-quarter-length shot of one adult on a clean studio backdrop, white by default, with light grey and a soft gradient as alternatives. That backdrop is generated fresh, never cut out of your picture. The prompt tells the model to take identity from your photographs, face, bone structure, skin tone, hair and build, while ignoring the clothing, pose and background in them.

One caveat before you upload your own face. The same prompt carries a standing instruction that the person must be an original character, not a real, identifiable individual, which keeps celebrity and franchise likenesses out. What comes back is a likeness in your direction, not a forensic match, and how close it lands varies run to run. Regenerate until it is a face you will stand behind. If it never quite reads as you, treat the character as an invented person built in your image, and tag it accordingly.

Does the same face come back every time?

Once the portrait exists, no downstream image ever sees your uploads. They stay on the character so a portrait regeneration can use them again, but every generated image points at the portrait instead. That is the mechanical reason the face holds still across a gallery, rather than a different stranger turning up in every frame. Eight images then read as one set.

The character is saved on your account and reused across generations. If you have set a default, the Enhance modal arrives with it already selected, next to a checkbox that takes it back out. The gallery flow features one character at a time, while the pickers on single-image regeneration, campaigns and the chat editor let you pick two. Marketing ads and campaign posts read the same anchor on Starter and above, and those surfaces are newer, so check what comes back.

Picmato then checks the rendered face against that anchor. The check runs on the seven slots that carry a character by default, plus any slot where you ticked Add to this image. Identity QA records not run, passed, failed or unavailable, and slots outside the check record not run. A failed check buys a free re-render. If the retry fails hard as well, with the character missing, duplicated or replaced by a stranger, the image is withheld and its credits refunded. A softer miss publishes the stronger attempt with a quality warning.

Where can a brand character appear?

Seven of the 45 shot types include a character by default: Lifestyle, Problem vs Solution, Action Shot, Before & After, Scale Reference, User Scenario and On Model. Seven is the default, though, not a ceiling. How much of the person appears depends on the frame. Full scene puts the whole character in the shot, partial contact puts only hands and forearms in frame, and panel-adjacent keeps the character beside a panel.

When you regenerate a single image, a checkbox called Add to this image appears once a character is selected and the slot does not use one by default. That is an explicit override, and it applies to that one image. Single-feature shots route differently, because there the feature treatment decides. Result state, single-step use, fit with a counterpart and styled versatility admit a person on their own. Macro surface proof, mechanism in action, quantity made visible, printed mark enlarged and component isolated stay product-only and refuse the override.

Where the frame is meant to hold a character, meaning the seven default slots and any slot you overrode, it has to be doing something with the product. Merely standing near it is treated as a failed image. Where a feature treatment only permits a person, the character may appear, but nothing demands it and no identity check runs on that frame. Two characters is the maximum in one generation, and a frame that calls for both has to give them one shared moment.

Amazon's tag, and the case nobody has ruled on

Amazon announced in late July 2026 that photorealistic AI-generated people must be declared in the file itself. The IPTC keyword contains-synthetic-performer goes in the XMP dc:subject field, before upload. It covers product images, video and A+ Content, in any of Amazon's worldwide stores. Amazon says it will then add an indicator, where applicable, telling customers the image includes AI-generated people. It has not described that indicator. Already-published content is grandfathered, so the keyword matters on your next upload.

An invented brand character is plainly inside the rule. A character generated from photographs of you is the hard case. Amazon exempts content that only features real people, even if AI has altered them, and that wording is written for an edited photograph. A generated likeness is not that, and it is not an invented stranger either. Amazon has published no clarification, and no court has decided whether the rule reaches a generated likeness of a real, consenting person.

The EU is not silent in the same way, but Amazon governs the file you are about to upload. So tag it. Tagging when you did not have to costs you a keyword. Not tagging when you should have is a policy breach on every listing carrying the image. Amazon lists four exemptions, and if any one describes your file the rule does not reach it.

• ✓Content that only features real people, even if they have been altered with AI tools
• ✓Content that only features characters from movies, TV programs, streaming content, documentaries, video games, or other expressive works
• ✓Content that does not feature any people
• ✓Content that does not feature photorealistic people

New York and the EU point in different directions

New York's synthetic performer law took effect on 9 June 2026. A synthetic performer is a digitally created asset meant to look like a human performer who is not recognisable as any identifiable natural performer. A character that is recognisably you probably falls outside that, and lands under consent and likeness rules instead. An invented character is the core case the statute was written for.

The duty is to disclose conspicuously in an advertisement, where the advertiser has actual knowledge. The penalty is $1,000 for a first violation and $5,000 for any after that. Whether a still listing image counts as an advertisement on its own is unsettled. Note the direction of travel. On this reading your own rendered face is the safer one, and that is the opposite of where our reading of Amazon's exemption lands.

Article 50 of the AI Act has applied since 2 August 2026, and it splits the work in two. Machine-readable marking of synthetic output under Article 50(2) is the duty of whoever provides the AI system, not yours. Providers already on the market get until 2 December 2026. It also spares assistive standard editing that does not substantially alter the input. What lands on you is Article 50(4): as a deployer, disclose where a person could take the content for real. Practitioners read the Commission's guidelines as treating realistic AI avatars and digital replicas of real people as deep fakes.

Who adds the tag, you or the tool?

Picmato writes no IPTC or XMP keywords at any point. Reference photos you upload are re-encoded to WebP on the way in, which drops whatever metadata they carried. Generated files come back as the provider returned them: Picmato adds no keywords of its own and strips nothing the image model wrote. So the keyword has to be added after download, in an IPTC-capable editor, by you. No image generator writes it, ours included. The one exception is not a generator. Amazon's Creative Assets uploader in A+ Content Manager has an AI-generated people checkbox, and applies the metadata to those assets for you.

One more thing a tag will not fix. BBB National Programs makes the point that a disclosure label does not rescue a message that is untrue. If a generated face implies a customer testimonial that never happened, labelling the image AI-generated does not repair the claim underneath. None of this is legal advice, and none of it is checked by Picmato. We do not verify a likeness release, confirm that anyone agreed to be used, or check a file against a marketplace's rules. Where a real person's likeness is involved, get the consent in writing and have a lawyer read it.