NaiLIA - Multimodal Nail Design Retrieval Based on Dense Intent Descriptions and Palette Queries

Sep 29, 2026

Upload by Connor

NaiLIA addresses multimodal retrieval of nail design images from dense intent descriptions that capture layered user goals, including painted elements, decorative embellishments, themes, and overall impressions, plus optional palette queries expressing zero to multiple colors with subtle continuous nuances. Existing vision-language models struggle to jointly model such unconstrained attributes and fine-grained color intent. NaiLIA introduces relaxed alignment via confidence-based loss for unlabeled positive images, aligning retrieval scores to the provided description and palette. Evaluation uses 10,625 culturally diverse images annotated by 200+ annotators, showing NaiLIA outperforms standard methods.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages23

Reading Time58 min

AccessFree to download

Introduction

Task motivation: dense intent and palette queries

Challenges with existing vision-language models

Proposed method: NaiLIA

Main contributions

References

NaiLIA: Multimodal Nail Design Retrieval Based on Dense Intent Descriptions and Palette Queries Kanon Amemiya Daichi Yashima Kei Katsumata Takumi Komatsu Ryosuke Korekata Seitaro Otsuki Komei Sugiura Keio University arXiv :2603 .05446v1 [ cs .CV] 5 Mar 2026 Abstract We focus on the task of retrieving nail design images based on dense intent descriptions, which represent multi-layered user intent for nail designs. This is challenging because such descriptions specify unconstrained painted elements and pre-manufactured embellishments as well as visual characteristics, themes, and overall impressions. In addition to these descriptions, we assume that users provide palette queries by specifying zero or more colors via a color picker, enabling the expression of subtle and continuous color nuances. Existing vision-language foundation models often struggle to incorporate such descriptions and palettes. To address this, we propose NaiLIA, a multimodal retrieval method for nail design images, which comprehensively aligns with dense intent descriptions and palette queries during retrieval. Our approach introduces a relaxed loss based on confidence scores for unlabeled images that can align with the descriptions. To evaluate NaiLIA, we constructed a benchmark consisting of 10,625 images collected from people with diverse cultural backgrounds. The images were annotated with long and dense intent descriptions given by over 200 annotators. Experimental results demonstrate that NaiLIA outperforms standard methods. Our project page, code, and dataset are available at: [https://nailia-94dpr.kinsta.page/](https://nailia-94dpr.kinsta.page/). 1. Introduction The global nail salon market is valued at approximately USD 11 billion, indicating a significant demand for the ability to search for nail designs meeting user preferences and/or nail artists capable of applying these designs. When customers request nail designs from nail artists, they often articulate their preferences by specifying painted elements (e.g., patterns) and decorative elements (e.g., nail accessories), as well as the overall themes and impressions that emerge from the interplay of these elements. It would be beneficial if the customers could search for nail design images using such dense intent descriptions. Although one can imagine that a simple text-to-image Figure 1 . A typical use case for our task. The user inputs a dense intent description and an optional palette query. The palette query allows the user to select zero or more colors through a color picker interface. The model should rank the leftmost image higher than the middle image (painted in a darker purple than specified by the palette query) and the rightmost image (featuring realistic shell ornaments rather than the intended shell-inspired design) . Details of this task are provided in Section 3.1. approach would suffice, retrieval-based approaches using real-world images are more practical. Indeed, nail artists have reported difficulties when clients present AI-generated references because such images often violate physical constraints (e.g., unrealizable nail accessories), making faithful reproduction difficult or infeasible. Moreover, several beauty service platforms have begun restricting the use or posting of AI-generated images. In this study, we focus on the task of retrieving nail design images based on dense intent descriptions. The descriptions represent user intent for nail designs in a detailed and multi-layered manner (e.g., colors, patterns, themes, and overall impressions) . We assume that users can optionally specify colors (a palette query) via a color picker, enabling the expression of subtle and continuous color nuances. Fig. 1 shows a typical use case for our task. Given the description “I’d like a mermaid-themed, ... I want it to have a dreamy look with a gradient effect,” and the palette query consisting of (\#b2f7e1) and (\#de8ffc), the model should retrieve the leftmo

FAQs

What does NaiLIA retrieve and what inputs does it use?

NaiLIA retrieves nail design images using dense intent descriptions and an optional palette query that specifies zero or more colors.

Why is retrieval with dense intent descriptions challenging?

Descriptions include unconstrained painted elements, pre-manufactured decorative embellishments, abstract themes, spatial relations, and overall impressions, which are harder to match to concrete visual elements.

How does NaiLIA improve over standard vision-language retrieval methods?

NaiLIA estimates confidence scores for unlabeled positives and uses a confidence-based relaxed alignment module, avoiding issues like treating all non-positive samples as negatives and better handling continuous color nuances.

Sep 29, 2026

Upload by Connor