Interactive Machine Learning for Generative Models

Effective control of generative media models remains a challenge for specialised generation tasks, including where no suitable dataset to train a contrastive language model exists. We describe a new approach that enables users to interactively create bespoke text-to-media mappings for arbitrary media generation models, using a small number of examples. This approach - very distinct from contrastive language pretraining approaches - facilitates new strategies for using language to drive media creation in creative contexts not well served by existing methods.