How it Works

Your references are read by a person, not fed to a machine. The moodboards you send, the casting guidelines, the brand lines, the old campaign images: an operator studies all of it on screen and writes a description from it. That written description is the only thing that enters the generator. No reference image, no photograph, no likeness of any real person is ever uploaded, referenced or processed.

This is the rule the whole process is built on, and it is the reason the model that comes out belongs to nobody.

The operator writes in language a casting director would recognise and a lawyer would not object to: build, proportion, hair, skin tone, the quality of a jawline, the way a shoulder sits. Generic descriptors throughout. No names, no likeness references, no targeting of any individual.

If the brief cannot be written without naming someone, the brief does not get written. That constraint is not a limitation on the work, it is the work. A description precise enough to produce a specific face, and general enough that the face is not anybody's.

The brief enters the pipeline as text and parameters. The underlying models are fine tuned on product imagery only, 3D renders, flat lays and still life. There is no real person training data anywhere in the pipeline.

Every candidate is then verified before it goes any further:

Against your own references. The candidate face is compared visually against every face in the moodboards and casting materials you supplied. If it resembles one of them, it is rejected.

Against our catalogue. The candidate is checked for similarity against every identity Sartiq has already produced, so two brands never end up sharing a face.

Against the internet. The portrait is run through reverse image search on Google, Bing and Yandex. If a real person surfaces who looks like the candidate, the identity is rejected and regenerated.

No exceptions and no overrides. A candidate that fails any of the three does not proceed.

Brief. Before anything starts you provide a moodboard with at least five high resolution references, plus gender, age range, build, hair and skin tone. If one of these is missing the work does not begin, and we tell you which one. You can request a call to add anything else.

Exploration. Within one working day of a complete brief you receive ten candidate models, each with a portrait and a full length shot in basic garments. A second round of ten is included: tell us which of the first ten to work around, and you receive ten more within a working day.

Choice. You pick your candidate from the explorations.

Refinement. Five rounds are included. In each round you request one change and receive three versions within a working day. Choosing one of the three is included. A smaller nose and a heavier build are two changes, so they are two rounds, taken in sequence.

Approval. You approve the model and it enters your cast. Approval is nominal: the record shows who approved and when. Adding the model to a Location is completed within two working days of the request.

Refinements are recorded one trait at a time, with the operator who made the change and the named approver who signed it off, until the identity is locked.

At lock, the identity is delivered with an Identity Dossier: the references studied, the trait brief as written, the three verification results, the full iteration log, the named approvers on both sides, and the underlying models and data processors involved. It travels with the identity and is updated if the identity is ever refined.

The law does not require this. We produce it because the day a published image is questioned, a brand needs contemporaneous evidence rather than a reconstruction assembled after the fact.

Every image leaving the pipeline carries machine readable provenance with no configuration on your side: C2PA Content Credentials, EXIF and XMP metadata, and an imperceptible watermark. Three layers, on every file, so the image can be identified as synthetic wherever it travels and whatever strips its metadata along the way.

That covers the marking duty at our end. The visible disclosure on your published pages remains yours, and we supply the technical means for it, including rendering a visible label directly onto delivered files before they reach your publishing pipeline.

A recurring identity used across a campaign is a recurring identity, not a character with a life. It is never captioned as a real person, never given biographical detail, and never presented as an endorser making first person claims about a product.

That editorial discipline lives on your side, but you do not start from zero. The dossier is the documentary basis for the position that the face belongs to no one, verified against your own references, against our catalogue and against the open web, before a single garment was ever put on it.