Machine learning-generated imagery offers new ways to visualise the Anthropocene beyond documentation. While climate crisis imagery—red skies, bleached coral, receding glaciers—signals disaster, it often arrives too late, depicting past events. This chapter explores how image-based machine learning, trained on vast datasets, can reframe environmental collapse through speculative futures. Examining artistic projects like Refik Anadol’s Quantum Memories and Anna Ridler’s The Shell Record, we argue that machine learning can expand perception, disorient expectations, and inspire action, paralleling the speculative nature of climate data modelling.