The realm of generative AI is witnessing rapid advancements, with 2023 marking a significant stride in the domain. Meta, formerly Facebook, has introduced Emu, a groundbreaking foundational model for image generation, at this year’s Meta Connect event. This technology underpins numerous AI experiences across Meta’s app family, notably in Instagram’s AI image editing tools. These tools enable users to transform photos by altering their visual style or background. Moreover, the Imagine feature in Meta AI facilitates the generation of photorealistic images within messages or group chats.
Breakthroughs in Video Generation: Emu Video
Emu Video emerges as a pivotal development, utilizing the Emu model for text-to-video generation. This innovative approach, based on diffusion models, offers a simple yet efficient method for creating high-quality videos. The process involves two phases: initially generating images from text prompts and subsequently creating videos conditioned on both text and images. This factorized methodology allows for efficient training of video generation models. Emu Video’s superiority is evident, as it only requires two diffusion models to produce 512x512 videos at 16 fps, a stark contrast to previous methods requiring multiple models. Human evaluations have shown a strong preference for Emu Video, with its performance outshining previous technologies in both quality and adherence to text prompts.
Revolutionizing Image Editing: Emu Edit
Meta’s Emu Edit represents a paradigm shift in image editing, focusing on precise pixel-level alterations. This tool enables intricate editing tasks such as local and global modifications, background adjustments, and color and geometric transformations. Emu Edit
Read more on blockchain.news