2.09 Get inspired: Reimagining the past with AI image generation

From preservation to creative reconstruction

The previous modules have established a blueprint for a heritage project focused on ethically preserving and presenting collected materials. Now, we explore a more creative avenue that builds upon these skills: using generative AI to address gaps in the visual historical record. What happens when a precious old photograph is torn, faded, or only shows part of a scene? Generative AI offers powerful tools to help us fill in these gaps, not as a replacement for historical fact, but as a method of creative and informed reconstruction. While Google Gemini has been our versatile assistant for text and analysis, for specialised creative media generation, we will now turn to dedicated platforms designed specifically for creating images, music, and video. Do note that Google Gemini is a multimodal model, meaning it does have the capabilities to produce images, sound, and video; however, in the following sections and modules we will also introduce some other AI tools and models.

AI Inpainting for historical photographs

The technique of using AI to fill in missing parts of an image is often called inpainting or, more recently, is a feature of tools with generative fill capabilities.

Tools like Adobe Firefly are particularly well-suited for this, as they are trained on a library of licensed, commercially safe content, which helps mitigate some copyright and ethical concerns.

These tools allow a user to select a missing or damaged area of an image and describe in text what should be there. The AI then generates a new section that seamlessly blends with the existing photograph.

This technique has profound implications for cultural heritage. Researchers have already used sophisticated AI to create detailed 3D reconstructions of ancient reliefs from single, flat historical photographs, digitally revealing details lost to time.

This same principle can be applied to restore damaged personal or community archives.

AI demonstration: Completing a historical photograph

Let us imagine that during the HER[AI]TAGE project, we found a single, damaged photograph from the 1950s showing a traditional fishing boat on the Danube. The photo is torn, and the back half of the boat is missing. However, we have detailed oral history descriptions of what these boats looked like. We can use this information to guide the AI.

Prompt for an AI image generator with Generative Fill (e.g., Adobe Firefly):

This is a partial photo of a 1950s wooden fishing boat on the river Danube. The back half of the boat is missing. Based on the existing part of the image, and the following description from an oral history interview, please complete the photograph in a photorealistic style: “The boat was a traditional ‘chaika,’ long and narrow, made of dark oak wood, with a high, curved stern. It would have had simple fishing nets piled in the back and a small, carved wooden tiller for steering.”

The AI would analyse the existing pixels, textures, and lighting of the original photo and use the text description to generate the missing portion. The result would be a complete image that visually represents the historical artefact more fully than the damaged original.

Demonstration of in-painting. Image by rawpixel.com.

Ethical considerations: Reconstruction vs. reality

It is crucial to approach this kind of creative reconstruction with transparency. The AI-completed image is not a historical document; it is an interpretation based on the available evidence (both visual and oral). When presenting such an image in an exhibit or publication, it must be clearly and explicitly labelled as an “AI-assisted reconstruction”.

This ensures that viewers understand the distinction between the original artefact and the creative addition. This practice respects historical integrity while still allowing us to use technology to create a more complete and engaging visual experience. By combining fragmented historical data with AI’s creative power, we can offer new ways to see and connect with the past.