Two young characters, one blonde, one brunette, cast from roughly 200 generated faces and styled for Barbie at Boyner. A careful, age-accurate kids campaign produced entirely with AI.
Barbie, part of the Boyner family in Turkey, needed a kids campaign that read as photographed, not generated: two young girls, styled and shot with the warmth and precision of real children's editorial.
Kids were the hardest brief we had taken on. In 2025, generating children with AI was a genuine frontier. Proportions drift, faces age a few years between frames, and the smallest error in scale or expression reads instantly as wrong. Accuracy mattered more here than in any adult campaign.
So we generated wide and cast narrow. From roughly 200 generated faces we selected two girls, one blonde and one brunette, whose features held up across angles, lighting and expression, and who read as real children a parent would recognise.
From there we built the lookbook. Each outfit was styled garment-first, with scale and age treated as hard constraints at every step. A waistline, a sleeve length, the size of a hand on a collar, all of it had to stay true to a child's body rather than a shrunken adult's.
FLUX dev carried the LoRA training and upscaling. Final accuracy and consistency came from creative review: targeted fixes in NanoBanana and Photoshop, frame by frame, until every detail held across the full campaign.
One film for each girl and one with both of them, the same two characters in motion.
We generated roughly 200 faces and selected two, one blonde and one brunette, as the campaign's locked leads.
Three final looks per girl, plus the shot of both together, selected from a much larger set generated in the backend.
The directed range behind the selects: framing, distance and body language explored per outfit, both girls together and alone.
We treated scale and age as hard constraints at every stage, generating a large set, casting two leads, and finishing the details by hand.
Roughly 200 child faces generated, then narrowed to two, one blonde and one brunette, chosen for features that stayed consistent across angles and read as real children rather than miniature adults.
FLUX dev carried the LoRA training and the upscaling. Kids demanded tighter datasets: every pass tagged for age, scale and proportion, not just pose and lighting, so the body never drifted older or larger.
Where the model drifted, creative review caught it. Targeted fixes in NanoBanana and Photoshop kept scale, age and continuity true across every frame of the campaign.
Blonde · Portrait
Hero portraitFinal select
Blonde · Detail
Styling & scale detailFinal select
Brunette · Portrait
Hero portraitFinal select
Brunette · Detail
Styling & scale detailFinal select
“ A campaign is no longer something you only shoot. AI gets you most of the way, and the last ten percent, scale, age, the details a parent would notice, is still creative work, done by hand.ILLUSORR
