Consent-trained diffusion twin with pose/wardrobe/location style packs
Commissioned by a luxury beauty brand based in the US (NDA). The brief was to create a digital twin of the campaign’s lead model (“Monika”) so the team could generate new, on-brand images without constant reshoots.
I captured a base dataset with full consent and trained a diffusion model on the model’s likeness. On top of that, I built style packs for specific body positions, locations, actions and clothes, so creative could ask for precise variations—“studio, three-quarter, soft key, evening dress,” or “street, walking, daylight, denim”—and get consistent, identity-true results.
The system is designed for privacy and control: trained on anonymised, rights-cleared material; scoped to approved uses; and tuned to keep skin, hair and expression natural rather than “AI-slick.” In practice, it means faster iteration, reliable continuity across channels, and fewer pickups—while the final images look like they were simply shot well.
This work is under development and strictly confidential.
I captured a base dataset with full consent and trained a diffusion model on the model’s likeness. On top of that, I built style packs for specific body positions, locations, actions and clothes, so creative could ask for precise variations—“studio, three-quarter, soft key, evening dress,” or “street, walking, daylight, denim”—and get consistent, identity-true results.
The system is designed for privacy and control: trained on anonymised, rights-cleared material; scoped to approved uses; and tuned to keep skin, hair and expression natural rather than “AI-slick.” In practice, it means faster iteration, reliable continuity across channels, and fewer pickups—while the final images look like they were simply shot well.
This work is under development and strictly confidential.
