Meta is working on video generation for its standalone Meta AI version. Currently, in addition to text conversations, users can click on the Imagine button for image generation, and find inspiration from highlighted prompts on the Imagine page. It appears that in the future, users will also be able to generate videos directly and combine them into bigger stories.
Here, instead of images, users see highlights of other videos and prompts. After entering a video prompt or selecting a highlighted one, a progress page appears with multiple placeholders for video generation. These videos are vertical, making them suitable for Instagram Reels, indicating that this feature might be suitable for generating AI videos for the Reels platform.
Highlighted prompts may contain different scenes, which are broken down into individual video files. If a prompt involves a single object, this object appears across all the videos. For more complex scenes, different scenes will appear in separate videos. This approach is similar to Google's storyboard method for video generation, but with Meta, users describe everything with a prompt, generating multiple video shots that can be combined later. Currently, combining them into a single video requires downloading each video separately and merging them manually.
Additionally, hints in the codebase suggest that the LLama 3 405B model may soon be available, potentially exclusive to premium plans. The Premium offer for Meta AI, rumoured by The Information, seems to be in the code and may coincide with the release of the LLama 3 405B model, scheduled for July 23.
This suggests that tomorrow may bring several announcements related to Meta AI products. Furthermore, the code mentions AI Studio, which could be a separate product targeting developers. It is possible that Meta AI will offer solutions similar to OpenAI Playground or Google AI Studio, allowing users to try Meta AI models and integrate them seamlessly into their own products.