Meta announced a new AI video generator tool called Make-A-Video today that can create content on the fly with image or text prompts, similar to existing software like Stable Diffusion and DALL-E. It can even take existing videos and create variations. It is currently unavailable to the public.
On the official press release page, Meta shows off examples of videos generated entirely from text. Several examples include a “teddy bear painting a portrait” and a “cat watching TV with a remote in hand.”
The presentation also showed off the ability of the AI generator application to animate static source images. For instance, a still picture of a turtle that is processed through the AI model can be animated to look like it’s swimming.
This tool arrives sooner than the industry expected because it builds off existing image generation work from other frameworks like DALL-E from OpenAI.
Earlier this year, in July, Meta announced its own solution for text-to-image known as Make-A-Scene.
Rather than training the model for Make-A-Video on video data with captioned descriptions, Meta opted for image synthesis data and unlabeled training data for videos to allow the model to get a sense of where an image or text prompt may exist. It then predicts the outcome of an image and displays a scene in motion over a brief period.
Currently, there is no announcement with respect to the availability of Make-A-Video to the public or the individual that would have access. However, a sign-up form has been provided for people to fill out if they would like to try it out at some point in the future.
Meta is aware that the power to create on-demand videos with photorealistic quality presents many social hazards. Watermarks are included on all AI-generated content using Make-A-Video are put in place. This is to “help ensure viewers know the video was generated with AI and is not a captured video.”