Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Moonworks Lunara Aesthetic Dataset

Yan Wang, M M Sayeef Abdullah, Partho Hassan, Sabit Hassan

TL;DR

The Lunara Aesthetic Dataset addresses the lack of openly reusable, high-quality data for aesthetic prompting in text-to-image generation. It introduces 2,000 Lunara-generated image–prompt pairs with human-refined prompts and structured annotations across regional and media styles, released under Apache 2.0. Quantitative evaluations show substantially higher aesthetic scores than baselines such as CC3M, LAION-2B-Aesthetic, and WIT, along with robust image–text alignment and cross-modal retrieval. The dataset enables reproducible research on prompt adherence, style conditioning, and diagnostic evaluation for modern image-generation systems, with plans to broaden semantic variation and regional granularity in future releases.

Abstract

The dataset spans diverse artistic styles, including regionally grounded aesthetics from the Middle East, Northern Europe, East Asia, and South Asia, alongside general categories such as sketch and oil painting. All images are generated using the Moonworks Lunara model and intentionally crafted to embody distinct, high-quality aesthetic styles, yielding a first-of-its-kind dataset with substantially higher aesthetic scores, exceeding even aesthetics-focused datasets, and general-purpose datasets by a larger margin. Each image is accompanied by a human-refined prompt and structured annotations that jointly describe salient objects, attributes, relationships, and stylistic cues. Unlike large-scale web-derived datasets that emphasize breadth over precision, the Lunara Aesthetic Dataset prioritizes aesthetic quality, stylistic diversity, and licensing transparency, and is released under the Apache 2.0 license to support research and unrestricted academic and commercial use.

Moonworks Lunara Aesthetic Dataset

TL;DR

The Lunara Aesthetic Dataset addresses the lack of openly reusable, high-quality data for aesthetic prompting in text-to-image generation. It introduces 2,000 Lunara-generated image–prompt pairs with human-refined prompts and structured annotations across regional and media styles, released under Apache 2.0. Quantitative evaluations show substantially higher aesthetic scores than baselines such as CC3M, LAION-2B-Aesthetic, and WIT, along with robust image–text alignment and cross-modal retrieval. The dataset enables reproducible research on prompt adherence, style conditioning, and diagnostic evaluation for modern image-generation systems, with plans to broaden semantic variation and regional granularity in future releases.

Abstract

The dataset spans diverse artistic styles, including regionally grounded aesthetics from the Middle East, Northern Europe, East Asia, and South Asia, alongside general categories such as sketch and oil painting. All images are generated using the Moonworks Lunara model and intentionally crafted to embody distinct, high-quality aesthetic styles, yielding a first-of-its-kind dataset with substantially higher aesthetic scores, exceeding even aesthetics-focused datasets, and general-purpose datasets by a larger margin. Each image is accompanied by a human-refined prompt and structured annotations that jointly describe salient objects, attributes, relationships, and stylistic cues. Unlike large-scale web-derived datasets that emphasize breadth over precision, the Lunara Aesthetic Dataset prioritizes aesthetic quality, stylistic diversity, and licensing transparency, and is released under the Apache 2.0 license to support research and unrestricted academic and commercial use.
Paper Structure (14 sections, 6 figures, 4 tables)

This paper contains 14 sections, 6 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Dataset Overview
  3. Prompt Characteristics
  4. Topic Distribution
  5. Regional and Style Distribution
  6. Intended Use
  7. Dataset Creation and Annotation Pipeline
  8. Dataset Evaluation
  9. Aesthetic Preference Analysis.
  10. Image--Text Alignment
  11. Cross-Modal Retrieval
  12. Visual Diversity
  13. Discussion
  14. Conclusion and Future Releases

Figures (6)

  • Figure 1: Distribution of key topics in the Lunara Aesthetic Dataset.
  • Figure 2: Style distribution of the Lunara Aesthetic Dataset across four geographical regions as well as region-agnostic category.
  • Figure 3: Overview of the Lunara image generation and annotation pipeline, illustrating model-based generation followed by human prompt refinement and filtering.
  • Figure 4: Regional style conditioning examples using a shared prompt across multiple cultural aesthetics.
  • Figure 5: General artistic style and medium conditioning examples using the same prompt.
  • ...and 1 more figures