Scenic is a code library dedicated to computer vision research based on attention models. It offers optimized training and evaluation loops, baseline models, and more, suitable for image, video, and audio multimodal data. Providing SOTA models and baselines, Scenic supports rapid prototyping and is free to use.