Micaelle Lages

Fashion + AI

FASHION + AI

RESEARCH

EXPERIMENTATION

AI Data Training for VOGUE UK (2023)

As a creative technologist freelancer, I worked at DE-YAN at an specific project focused on bridging the fashion industry and AI latest advancements. Generative AI was rising and VOGUE UK wanted to release a fully AI generated edition while still highlighting real designed pieces by large brands and real famous models. 

Our solution was to work on a personalized model trained on a large dataset with older editions of VOGUE UK. 

More than 100 editions of the magazine released since 2014 were unraleved by our AI team during a period of 2 months. Each editorial image was uploaded to our system and followed a very detail system:

Each piece of clothing and accessory was segmented in the image and received a specific tag with the designer’s brand. Models and actress’ faces were also segmented for the database accuracy. 

Ultimately the idea was for the annotated dataset to be used to training a personalized generative AI model to autonomously generate novel, stylistically consistent VOGUE UK fashion photoshoots.

GenAI and Fashion Experimentation (2024-25)

Over the last year, models are evolving in an impressive rate, allowing for capabilities such as face consistency and garments fidelity. Lately I have been dedicated to research the best models and approaches to translate clothes from one picture to another, testing on posing replication, virtual try-on of clothing and synthetic generated fashion editorial photography. 

The workflow utilizes FLUX models for high consistency of person faces, posing and artifacts.

The examples shared here take advantage of some of the segmentation done for VOGUE UK and displayed above.

Posing and face consistency

Posing reference and clothing consistency