Moonworks Lunara Aesthetic II: An Image Variation Dataset

TL;DR

Moonworks Lunara Aesthetic II provides a publicly released, ethically sourced dataset of $2854$ anchor-linked image variation pairs derived from $336$ originals to study identity preservation under contextual edits. It frames controlled transformations across six axes—illumination-time, weather-atmosphere, scene composition, color-tone, mood, and viewpoint—while maintaining identity and high aesthetics. The work introduces a diffusion-mixture Lunara model to generate variations and demonstrates strong identity stability ($4.65/5$) and high target-attribute realization ($87.2%$), with aesthetic scores surpassing several web-scale datasets. It offers a principled benchmark for contextual generalization, edit robustness, and relation-based supervision in image generation and editing, released under Apache 2.0 to support reproducibility and broad adoption.

Abstract

We introduce Lunara Aesthetic II, a publicly released, ethically sourced image dataset designed to support controlled evaluation and learning of contextual consistency in modern image generation and editing systems. The dataset comprises 2,854 anchor-linked variation pairs derived from original art and photographs created by Moonworks. Each variation pair applies contextual transformations, such as illumination, weather, viewpoint, scene composition, color tone, or mood; while preserving a stable underlying identity. Lunara Aesthetic II operationalizes identity-preserving contextual variation as a supervision signal while also retaining Lunara's signature high aesthetic scores. Results show high identity stability, strong target attribute realization, and a robust aesthetic profile that exceeds large-scale web datasets. Released under the Apache 2.0 license, Lunara Aesthetic II is intended for benchmarking, fine-tuning, and analysis of contextual generalization, identity preservation, and edit robustness in image generation and image-to-image systems with interpretable, relational supervision. The dataset is publicly available at: https://huggingface.co/datasets/moonworks/lunara-aesthetic-image-variations.

Figures (6)

• Figure 1: Distribution of topics in the image variation dataset.
• Figure 2: Conditional probability of variations co-occurring.
• Figure 3: Distribution of variation types in the image variation dataset.
• Figure 4: Contextual variations in original photographs. Each pair shows (a) and (b).
• Figure 5: contextual variations in original art. Each pair shows (a) and (b).