Phenaki is an AI model designed to produce videos of varying lengths from written text. Additionally, it has the capability to generate videos based on a single image and a prompt. This innovative video encoder-decoder model surpasses the existing per-frame baselines commonly utilized in research, excelling in both spatio-temporal quality and the number of tokens per video. Consequently, text tokens are transformed into video tokens through a bidirectional masked transformer that is conditioned on pre-computed text tokens. These generated video tokens are then de-tokenized to form the finalized video.