Phenaki is a model capable of generating realistic videos from textual prompts, with the ability to change prompts over time and create videos of any desired length. It uses a tokenizer with causal attention in time to compress videos into discrete tokens, and a bidirectional masked transformer to generate video tokens from text. Phenaki can generate videos conditioned on a sequence of prompts, making it the first model to study video generation from time variable prompts. It outperforms previous video generation methods in terms of spatio-temporal quality and token count per video.
Features
- Generates videos from textual prompts
- Allows for dynamic prompts that can change over time
- Can generate videos of any desired length
Use Cases
- Creating videos from text descriptions
- Generating videos for storytelling or visualization purposes
- Producing video content based on textual prompts
Suited For
- Content creators and storytellers
- Researchers in the field of video generation
- Anyone interested in generating videos from text prompts