The Gulf of Maine, one of the world’s most biodiverse marine regions, is warming faster than 99% of global oceans. A new MIT Sea Grant initiative, LOBSTgER – short for Learning Oceanic Bioecological Systems Through Generative Representations – is using generative AI and underwater photography to document and share this shifting ecosystem.
The project is co-led by underwater photographer and MIT visiting artist Keith Ellenbogen and PhD student Andreas Mentzelopoulos. It trains AI models on Ellenbogen’s real underwater images to generate scientifically accurate and emotionally resonant visuals. Unlike most generative models, LOBSTgER’s are custom-built using original code, avoiding outside data bias and ensuring ecological integrity.
AI-generated and AI-enhanced images replicate not just species and habitats but also artistic elements like light, color, and atmosphere. These visuals help overcome underwater photography challenges, such as poor visibility and fleeting encounters, while preserving biological accuracy.
LOBSTgER aims to deepen public engagement by combining scientific observation with artistic storytelling. The models can generate new marine life imagery or enhance real photographs to emphasize detail and clarity. The approach helps visualize species like blue sharks, ocean sunfish, and lion’s mane jellyfish, which are difficult to capture in ideal conditions.
Mentzelopoulos stresses the project’s goal is not to replace photography but to extend its reach. “We’re helping people see complexity in ways that are both intellectually rigorous and emotionally moving,” he said.
Still early in development, LOBSTgER envisions expanding to cover broader marine ecosystems worldwide – offering a new lens on conservation in a rapidly changing ocean.
