The rapid development of pretrained models, including large language and foundation models, opens up many new applications in geosciences. Fine tuning or deploying existing models directly to geoscience problems gives us all the benefits of very large models, trained with vast data sets, whilst reducing the compute cost and required specialist data sets, compared to older deep learning techniques. The potential cost and energy saving, and scientific power of these models can revolutionise AI in geosciences. In this session, we look forward to seeing state-of-the-art fine tuning techniques specialized for geoscience; novel deployment of pretrained models of any sort; and lessons learnt from trial and error. We would love to see applications across the geosciences, from geophysics to education, remote sensing to hands on field applications.