Phenaki is an AI tool that can generate realistic videos from textual prompts. It uses a new causal model to learn video representations and compress them into discrete tokens for variable-length videos. The tool can generate videos based on open-domain, time-variable prompts and performs better than per-frame baselines. It can also work with a smaller number of video-text examples and larger image-text corpora. It is the first time this type of video generation has been studied in a paper.