Researchers create AI tool to generate historical art style images

Researchers developed a system called The RenAIssance to turn collections of European historical artworks into a reusable resource for image generation.

The team used a pipeline to create scene-level captions for historical pieces and paired them with images from the DEArt collection. The researchers used these pairs to train a compact LoRA adapter for the Stable Diffusion v1.5 model.

This adapter teaches the AI to produce images that match the visual style of historical art while still allowing it to render new concepts. The tool shows that users can generate both period-specific subjects and modern ideas using a consistent historical visual language. The researchers found that the system improves how closely generated images match the DEArt reference distribution.

While this change slightly reduces how well the images match specific text descriptions, it allows for more controllable creation of historical-art media. This method provides a building block for interactive authoring and creating rich media experiences based on historical styles.