Introducing Muse Image and Muse Video

Meta has unveiled Muse, a new family of generative models designed to handle both image and video creation. By leveraging a transformer-based approach, these models demonstrate high performance in generating visual content while maintaining efficiency, offering a new way to bridge the gap between text-based instructions and high-quality media production.

Capabilities of Muse Image

Muse Image is designed to follow user instructions with high fidelity, ensuring that the generated output aligns closely with the provided prompts. Beyond simple generation, the model excels at precise editing, allowing users to modify specific elements within an image.
The model also supports composition from multiple references, enabling users to combine various visual inputs into a single, cohesive result. By drawing on social context from Instagram, Muse Image is equipped to understand and generate content that reflects a broad range of visual styles and trends.

Advancements in Video Generation

The Muse framework extends its generative capabilities to video through Muse Video. This model utilizes the same underlying transformer architecture to produce fluid and coherent video sequences.
By applying the principles of efficient generation used in the image model, Muse Video maintains visual consistency across frames. This approach allows for the creation of dynamic content that remains responsive to user input, marking a significant step forward in how generative AI can be applied to moving imagery.