# Genesis 01 — Testing AI Vision's Understanding of High Fashion Nuances

> Genesis 01 compares CLIP, SigLIP, and DINOv2 on fashion-image uncertainty, impostor detection, and collection cohesion.

- Canonical: https://research.caeliai.com/research/genesis-01
- Markdown: https://research.caeliai.com/research/genesis-01.md
- Publisher: Caeliai (https://caeliai.com), independent AI commerce research and advisory practice
- Official LinkedIn: https://www.linkedin.com/company/caeliai/
- Author: Kalan Peace
- Founder LinkedIn: https://www.linkedin.com/in/kalan-peace-2b2a18198
- Contact: contact@caeliai.com

---

Genesis 01 · Research paper

# Testing AI vision’s understanding of high fashion nuances.

Kalan Peace · Caeliai · Published in 2025

Also published as “A Vision Benchmark for Fashion AI”.

The first study compares how CLIP, SigLIP, and DINOv2 handle fashion images, focusing on uncertainty, impostor detection, and collection cohesion.

Using 12,147 Rick Owens runway images across 23 years, the study runs 3.66 million image comparisons. In this evaluation, SigLIP is the only model that becomes more cautious when the image is ambiguous.

**Quick read**

This page is the abstract version of the paper: scope, method, the result that matters, and the three benchmark scenes that explain why.

## What the study shows.

The study is less about raw similarity scores and more about behavior: whether a model can recognize a design family, stay coherent across a collection, and mark uncertainty when it should.

**Scope**

Genesis 01 compares three vision backbones across three tasks: impostor detection, collection cohesion, and exact look matching. The goal is not generic image classification. It is how the models handle uncertain fashion images.

- **3** - Models compared
- **3** - Benchmark tasks
- **2002-2025** - Temporal range
**Key result**

> In this evaluation, SigLIP is the only model that becomes more cautious when the image is an impostor.

CLIP and DINOv2 remain more confident on the ambiguous cases. SigLIP is more expensive, but it shows the strongest uncertainty behavior in this evaluation.

- **63.5%** - Collection purity
- **9.6x** - Higher processing cost
## The benchmark is built from three concrete scenes.

Each scene maps to a real product problem: false similarity, weak collection understanding, or failure to retrieve the precise object the user means.

**01 / Impostor detection**

From pixels alone, the model has to recognize that visually adjacent looks are still the wrong designer or wrong season and should trigger uncertainty.

[Image: Genesis 01 query image]

[Image: Genesis 01 impostor example one]

[Image: Genesis 01 impostor example two]

**02 / Collection cohesion**

One runway look should retrieve its family, not a group of vaguely similar silhouettes from different years. This separates collection matching from general visual similarity.

[Image: Genesis 01 family resemblance query]

[Image: Genesis 01 family resemblance match one]

[Image: Genesis 01 family resemblance contamination example]

**03 / Exact match**

Retrieval also has to stay precise. When a user points at a specific look, the system should find that exact target instead of drifting toward a general aesthetic neighborhood.

[Image: Genesis 01 exact match target]

[Image: Genesis 01 found match]

[Image: Genesis 01 haystack comparison image]

**Operational read**

> A vision model needs to show when it is unsure.

If the model cannot mark uncertainty, its output may favor a visually similar but incorrect reference. That is a model-behavior issue, not only a score.

**Why it matters**

Recommendation systems may use visual similarity to describe products. If a model cannot separate designer identity from visual similarity, its output may become an overconfident approximation.