AI at Meta

?? Today we’re excited to premiere Meta Movie Gen: the most advanced media foundation models to-date. Developed by AI research teams at Meta, Movie Gen delivers state-of-the-art results across a range of capabilities. We’re excited for the potential of this line of research to usher in entirely new possibilities for casual creators and creative professionals alike. More details and examples of what Movie Gen can do ?? https://go.fb.me/00mlgt Movie Gen Research Paper ?? https://go.fb.me/zfa8wf ??? Movie Gen models and capabilities ? Movie Gen Video: A 30B parameter transformer model that can generate high-quality and high-definition images and videos from a single text prompt. ??Movie Gen Audio: A 13B parameter transformer model can take a video input along with optional text prompts for controllability to generate high-fidelity audio synced to the video. It can generate ambient sound, instrumental background music and foley sound — delivering state-of-the-art results in audio quality, video-to-audio alignment and text-to-audio alignment. ??Precise video editing: Using a generated or existing video and accompanying text instructions as an input it can perform localized edits such as adding, removing or replacing elements — or global changes like background or style changes. ??Personalized videos: Using an image of a person and a text prompt, the model can generate a video with state-of-the-art results on character preservation and natural movement in video. We’re continuing to work closely with creative professionals from across the field to integrate their feedback as we work towards a potential release. We look forward to sharing more on this work and the creative possibilities it will enable in the future.

A few months ago, I took on the exciting challenge of crafting and delivering a Ted talk that shares my thoughts on how we can create an open, collaborative AI ecosystem, empowering researchers and communities worldwide to use AI in order to solve real-world problems. Credit for some content and all visuals to many collaborators at Meta! https://lnkd.in/e_ueu-7S

Llama has officially crossed 1 Billion downloads! To the global AI community of researchers, engineers, developers and hobbyists: We announced the first Llama models for the research community a little over two years ago and in that time your actions have spoken louder than words. Thank you for making it abundantly clear — a billion times over — that open source AI is how we'll create the next wave of world changing technologies, together. ????

New ads ranking research from Meta researchers. Compared to prior work, UKDSL can enable models to learn from a much larger set of unlabeled data, improving performance while being computationally efficient ?? https://go.fb.me/xrl0ja

Llama models were used to develop India's first open source audio language model — Shuka v1. The encoder-decoder model uses Sarvam's in-house audio encoder combined with Llama 3 as a decoder. More on this work ?? https://go.fb.me/se6aog Deep dive on Shuka v1 ?? https://go.fb.me/qirl9k

As Head of Business AI, Clara S. is dedicated to making AI accessible for companies of all sizes. At Meta, we envision a future where business AI can assist the hundreds of millions of small businesses using platforms like WhatsApp, Instagram and Facebook to connect with customers. Recognizing that not every business has the resources to develop bespoke AI agents or fine-tune LLMs, Clara believes Meta can leverage their scale and reach to provide small businesses with tools that were once available only to large companies with substantial resources. To learn more about Clara’s background and how she came to Meta, check out her interview with CNBC Changemakers: https://lnkd.in/gRx-B3WF ?? Clara’s team is hiring! Explore open roles in Business AI: Director, Product Management (Business Lead), Business AI https://lnkd.in/gWC7tCAs Product Manager, Business AI https://lnkd.in/gaqMe9cs Product Lead (Enterprise Foundations), Business AI https://lnkd.in/gEbgZvnr #LifeAtMeta #MetaCareers

AI is helping researchers and developers open up new avenues for cancer research and identify promising new therapies for patients. Read the full story ?? https://lnkd.in/gMnm-Xa5 Orakl Oncology trained our open source DINOv2 model on organoid images to more accurately predict patient responses in clinical settings. Their approach outperformed previous models specialized for organoids and is helping them accelerate their research.

New dataset from researchers at Meta — uCO3D, or UnCommon Objects in 3D, is the largest publicly-available object-centric dataset for 3D deep learning and 3D generative AI. More on this project ?? https://go.fb.me/8u86hq Documentation and download ?? https://go.fb.me/izrajn Highlights ? 170,000 videos depicting diverse objects from all directions. ??19.3TB of data. ??Objects come from the LVIS taxonomy of ~1000 categories, grouped into 50 super-categories. ??Full original videos instead of frames —?each annotated with object segmentation, camera poses and point clouds. ??3D Gaussian Splat reconstruction for each video. ??Long and short caption obtained for each scene with a large video-language model. ??Significantly improved annotation quality and size compared to previous datasets of its kind.

Our Meta PARTNR demo operationalizes human-robot collaboration. By sharing some of the challenges and insights, we hope to contribute to the development of the next wave of innovation in this space. Research paper ?? https://go.fb.me/prlg14? Dataset and code ?? https://go.fb.me/k9dzc0

Blended, an EdTech company in Germany, is using Llama 3.1 and Llama 3.2 models to enable a wide range of AI-native flows for personalized learning pathways, real-time feedback, instant educational content generation and social gamification. Their platform, built with Llama is unlocking greater productivity for educators, visual tutoring for students and increased access to quality education in regions impacted by crises ?? https://go.fb.me/259kub