The Google Research project Imagen Video (opens in new tab) is a text-to-video generator similar to Meta’s recently revealed Make-A-Video (opens in new tab) system. When prompted by the user, the tool is able to build looping video clips, and the results are as impressive as they are bizarre.
However, users eager to try out the tool will have to wait. Unlike Meta’s offering, which includes a sign-up option, the Google Research team is withholding public access for the time being, citing safeguarding issues.
Beauty in the AI of the beholder
This isn’t Google’s first rodeo in the AI-powered creative space. In May 2022, the team unveiled Imagen, a DALL-E-style AI generator capable of generating photorealistic still images based on written prompts.
The new AI-based video tool works the same way. Featuring what Google dubs “a high degree of controllability and world knowledge”, it can generate a diverse range of high-fidelity videos across different aesthetic styles, and with a deep understanding of 3D models. Creating the high-definition, 24fps videos is faster, too.
“With the help of progressive distillation, Imagen Video can generate high quality videos using just eight diffusion steps per sub-model. This speeds up video generation time substantially, by a factor of ~18x,” Jonathan Ho, one author of the research paper, tweeted (opens in new tab).
But before content marketing teams and social media creators get too excited about the prospect of easy video creation, it’s worth noting that Imagen Video is absolutely not ready for public consumption.
Despite filtering for “undesirable content”, the team admits “there is a risk that Imagen has encoded harmful stereotypes and representations”. So, until the AI – and its users – can be trusted, Google refuses to release Imagen Video without additional safeguards in place.
However, the reveal alone shows just how seriously companies are taking the technology, despite ongoing controversies surrounding artificial intelligence in digital art.