This document discusses creative AI and multimodality. It begins by looking at current possibilities for creative AI, including appropriating standard neural networks for creative use, reinforcement learning approaches that frame creativity as a game, recurrent neural networks, sequence-to-sequence models that treat creativity as translation, autoencoders, attention-based models, and generative adversarial networks. It also discusses needs for creative AI, including developing a system that marries a creative process with creative outputs using minimal human input data but with its own style and the ability for human-level supervision to enable rapid experimentation. The document frames creative AI as a "brush" that can be used for painting.