A vision benchmark for fashion AI.
The first study benchmarks fashion-intelligence behavior across CLIP, SigLIP, and DINOv2, focusing on uncertainty, impostor detection, and collection cohesion.
Genesis measures AI recommendation systems.
Genesis tracks how language models and vision systems perceive products, rank categories, and quietly shape commercial discovery.
This is not marketing collateral. It is an ongoing observation layer on top of current model behavior, recommendation logic, and shopping-interface drift.
Each report is a different lens on machine taste and machine perception.
Together they form a growing map of the systems now mediating product discovery, with a focus on what breaks, what gets privileged, and what brands can actually control.
LLMs’ favorites.
The second study maps how leading language models repeatedly converge on the same designers, garments, and fashion narratives when taste is inferred from prompts alone.
What these studies are really measuring.
Beyond raw outputs, the work is about the recommendation systems hidden inside current models: the assumptions they make, the aesthetics they overfit to, and the commercial surfaces they now influence.
Can a model actually distinguish brand, season, and design family from the image itself?
Which objects and creators recur across models when taste is generated from text?
How likely is a brand to be named in AI-mediated shopping?
Does the model know when it does not know, or does it hallucinate authority?
Genesis 03 expands the lens across categories.
The next report studies shopping behavior across multiple ecommerce verticals and compares how models prioritize brands under real consumer-style prompts.