![\"\"](\"https://scontent-sjc3-1.xx.fbcdn.net/v/t39.2365-6/752648857_903078942844172_5326165404058647623_n.png?_nc_cat=103&ccb=1-7&_nc_sid=e280be&_nc_ohc=E5SQsgj-s6cQ7kNvwFiOiny&_nc_oc=AdopW_lWY-KL52ypcqc42lwDkaLGVDvnZ6aJQD9Y26IMC6Rb_ZfxOJCQTUXeEiW-yEo&_nc_zt=14&_nc_ht=scontent-sjc3-1.xx&_nc_gid=tBbPS4ZQU1BKmBEn7Svf3A&_nc_ss=7b20f&oh=00_AQPvj0QCbv2d24hbDh7qxUOPriB3JyZGGafe5VXDksRdog&oe=6AEFCFB9\")
For people relying on assistive devices, every second counts. The unpredictable nature of everyday environments, such as a child darting across a sidewalk, the sudden appearance of a curb, or a dropped set of keys, requires an immediate reaction. Processing camera images and sensor data directly on the device, known as edge computing, empowers robotic mobility platforms to function as responsive tools. These tools must be robust and consistent across the wide range of dynamic environments in which people live.
\nAt the same time, deploying powerful AI models like DINOv3 and SAM on limited, battery-powered hardware presents significant engineering challenges. Real-world deployments must consider practical factors, including battery life, heat dissipation, unreliable network connectivity, and strict size and weight requirements.
\nHowever, overcoming these constraints mean nothing to the end user if they can't perform their activities of daily living using these new robotic systems. These newer methods allow users to interact with the robot more naturally, using their immediate surroundings as context, freeing both engineers and users from having to design and navigate complex, time-consuming interfaces. The ability to use natural language combined with image data to query the user’s and robot’s environment and provide more direct commands directly reduces the cognitive load and amount of context switching required for a user to do something as simple as picking up a cup off a table.
\nThis functionality is already being integrated by the RAMMP team into their first prototype. Leveraging tools built off of DINO to enable querying the robot’s image sensors to detect automatic door buttons, cups, and curbs/ground for navigation assistance. With this functionality now ready for real-world testing, engineers are focusing on voice and touch input, letting users select and interact with specific objects in their surroundings. In addition to the existing challenges of ensuring accuracy and temporal coherence of the model outputs, this provides the additional challenge of ensuring robustness and predictability across user prompts and inputs.
\nDINOv3 serves as a compact, efficient 'visual brain' for devices — a general-purpose foundation on which task-specific, lightweight modules can be layered for actions such as object detection or movement tracking, enabling reuse of visual data and conserving power.
\nApplying both models as part of the development of robotics systems, engineers optimize models for edge devices, reducing memory footprint, using lower precision when appropriate, and deploying in formats tailored for real-world conditions. This ensures reliable, real-time operation for users. By running at practical resolutions and with efficient batching, both models stay fast and dependable, even on the compact, battery-powered hardware used in robotic mobility platforms and robotic arms, — sometimes trading a little bit of boundary precision and/or feature detail for the speed and stability needed by users on the go. This balance between precision and practicality is central to the project's philosophy.
\n“For assistive robotics, performance is not measured by benchmark accuracy alone, but by whether a system can operate reliably in the unpredictability of everyday life," said Sivashankar Sivakanthan, Chief of Staff to the RAMMP project. “Running models like DINOv3 and SAM on-device is what enables real-time perception that users can trust - without relying on connectivity or compromising safety.”
\nRAMMP's perception system is built on RF-DETR, a lightweight detection model fine-tuned with DINOv2 embeddings. Training data is auto-labeled using SAM, enabling the team to rapidly generate high-quality annotations across the full range of angles, heights, backgrounds, and lighting situations that assistive devices encounter in the real world. Data augmentations and multi-view strategies further enforce consistency across perspectives. The result is a system that is smart, adaptive, and offers safer and more confident mobility.
\nBy combining SAM's labeling power with DINOv2's rich visual representations in a fine-tuned RF-DETR model, RAMMP achieves real-time 360-degree environmental awareness and adaptive object detection.
\n", "date_published": "2026-07-27T00:00:00+00:00", "external_url": "https://ai.meta.com/blog/assistive-robotics-university-of-pittsburgh-sam-dino/", "id": "urn:aggr:item:6ef9aab9dc2474d7d5df86e6ce37cdc88fffe2a1", "image": "https://scontent-sjc3-1.xx.fbcdn.net/v/t39.2365-6/753694960_2219191542190533_6830137062160879015_n.png?_nc_cat=105&ccb=1-7&_nc_sid=e280be&_nc_ohc=uN6SevXoH9kQ7kNvwHJ6P9c&_nc_oc=Adoy2HGf_mceuXR8Sm3zUYOTWJSSzy14rz1RIQYQq0TddftgUP8fX-6YHmI8jRRvIOs&_nc_zt=14&_nc_ht=scontent-sjc3-1.xx&_nc_gid=jKbf0X8OYPQM0-g56pCq4A&_nc_ss=7b20f&oh=00_AQMf7BsisR2R88jiKoDgCDlfH1YaJ8yk8R124qI8PGhDuA&oe=6AEFC820", "summary": "For people relying on assistive devices, every second counts. The unpredictable nature of everyday environments, such as a child darting across a sidewalk, the sudden appearance of a curb, or a dropped set of keys, requires an immediate…", "tags": [ "ai" ], "title": "Reimagining Independence: How Meta’s AI Models Are Helping the University of Pittsburgh Transform Assistive Robotics", "url": "https://aggr.aymericbeaumet.com/items/ai-meta-com/2026-07-27-reimagining-independence-how-meta-s-ai-models-are-helping/" }, { "content_html": "Lawrence Berkeley National Laboratory — one of the US Department of Energy's premier research laboratories, known for Nobel Prize-winning work in physics, chemistry, and materials science — operates some of the most advanced scientific facilities on the planet. Among them is the Advanced Light Source (ALS), a football field-sized facility that produces intensely bright beams of X-ray light, allowing researchers to study materials from the atomic and molecular scale all the way to plants. The ALS's instruments, known as beamlines, generate enormous quantities of data — and as recent facility upgrades have dramatically increased their resolution and speed, the volume of data has exploded beyond what scientists can keep up with.
\nThe numbers are staggering: The DOE's light and neutron source facilities now produce tens of petabytes of data annually — that's millions of gigabytes, roughly equivalent to streaming 2 million hours of HD video. This backlog didn't always exist. Upgraded detectors, which have gone from capturing a single image every six seconds to 100,000 images per second, mean these facilities now generate orders of magnitude more data than they did a decade ago, and traditional manual analysis simply can't keep pace.
\nThe problem goes beyond volume: domain experts are scarce and overwhelmed, and modern in-situ experiments — where scientists observe dynamic processes like chemical reactions or material failures as they occur — demand real-time interpretation that no human team can deliver manually.
\nMuch of the analysis challenge comes down to one task: segmentation — the process of identifying and drawing precise boundaries around distinct structures within an image. In computer vision, segmentation is what enables everything from medical scans that distinguish tumors from healthy tissue to autonomous vehicles that separate pedestrians from pavement. In scientific research, segmentation is what transforms a raw X-ray image from a wall of grayscale pixels into a labeled map of meaningful structures — cell walls, mineral grains, semiconductor layers — that researchers can quantify and compare across experiments.
\nIn late 2025, the White House launched The Genesis Mission, a sweeping national initiative to accelerate scientific discovery and technological leadership using advanced artificial intelligence, led by DOE. SYNAPS-I (SYnergistic Neutron And Photon Science – Intelligence) is one of its flagship projects: a multi-lab initiative led by Berkeley Lab, in partnership with Argonne, Brookhaven, Oak Ridge, and SLAC National Laboratories, aimed at transforming data analysis across X-ray and neutron science from a months-long bottleneck into a real-time discovery engine, with scientific imaging as a major target. Nowhere is that bottleneck more acute than in image segmentation, where extracting meaningful structures from experimental data can consume weeks of expert effort per dataset.
\nAt the heart of SYNAPS-I's segmentation pipeline are two open-source foundation models released by Meta: Segment Anything Model 3 (SAM 3) and DINOv3.
\nDINOv3 is a self-supervised vision model, meaning it learns visual patterns from raw images without requiring humans to label them first. It excels at understanding what different structures in an image represent and where they are located. SAM takes that understanding a step further, drawing precise boundaries around individual objects in an image — much like a scientist carefully outlining structures by hand, but in seconds rather than hours.
\nTogether, the two models form a complementary pipeline: SAM delivers precise, pixel-level boundaries, while DINO provides global context to identify each structure and its place within the sample. The SYNAPS-I team fine-tuned both models on scientific imaging data collected at DOE beamlines, then deployed them across 300 A100 GPUs — the high-performance computing chips that power today's most advanced AI systems — at national supercomputing facilities such as NERSC. The result: a fully reconstructed, semantically labeled 3D volume delivered back to the scientist physically standing at the beamline instrument, ready for interpretation while the experiment is still running. Total turnaround: approximately 15 minutes.
\nThe SYNAPS-I team demonstrated this pipeline on a pressing agricultural challenge — understanding how grapevines respond to drought at the cellular level. Using micro-CT scans collected at the Advanced Light Source, the pipeline reconstructs 3D volumes of vine stems and automatically identifies xylem vessels — the microscopic tubes responsible for water transport within the plant. By tracking how these vessels change as drought progresses, researchers gain insights that could inform the development of drought-resilient crops, and provide solutions for agricultural resilience into the future.
\n
Micro-CT scan of a grapevine stem, segmented by SYNAPS-I. Cyan: hydrated xylem vessels; dark purple: dry.
\nWhat previously required a month of expert annotation per time step now takes 15 minutes, enabling scientists to study dynamic biological processes at the speed of data acquisition itself.
\nNational laboratories keep prepublication research data and AI models on government infrastructure, not external cloud services. This work must be managed on secure platforms while in progress. Meta's open source approach makes this possible. The SYNAPS-I team can download, fine-tune, and deploy SAM and DINO within their own secure computing environments, adapting models originally trained on natural images to scientific domains they were never designed for.
\nWith 60 researchers across five national labs, SYNAPS-I is building toward a future where user facilities operate as intelligent discovery platforms — where AI doesn't just process data faster, but helps scientists generate hypotheses, recommends next experiments, and transfers knowledge across facilities so that a breakthrough at one beamline benefits researchers at all of them. As Genesis scales from seed projects to full programs, that open source foundation is poised to accelerate discovery across an expanding set of national priorities.
\nAt the recent Trillion Parameter Consortium, DOE Under Secretary Dario Gil referred to the promise of this effort.
\n"By seamlessly combining AI, advanced computing, and experimental systems, SYNAPS-I analyzes data as it's produced and guides experiments in real time, replacing slow manual steps with adaptive, automated decision-making," he said. "This compresses discovery time from days to moments and establishes a continuous, self-improving model of science that will be essential to realizing the full potential of the Genesis Mission."
\n", "date_published": "2026-07-21T00:00:00+00:00", "external_url": "https://ai.meta.com/blog/genesis-mission-lawrence-berkeley-national-laboratory-segment-anything-dino/", "id": "urn:aggr:item:7ecac91c875f80dc7e5fd474d1dc1a011eecc941", "image": "https://scontent-sjc3-1.xx.fbcdn.net/v/t39.2365-6/752631850_918439901272497_2117962010845605116_n.png?_nc_cat=103&ccb=1-7&_nc_sid=e280be&_nc_ohc=vFQSBsXzYkgQ7kNvwGvsnUV&_nc_oc=Adro6nsV7BQjCQRrJ4eHeNRkfVBtgg65UJ1eyvP-6Hb-Ee7kbUdMe8wjhZNQGiOlw-Q&_nc_zt=14&_nc_ht=scontent-sjc3-1.xx&_nc_gid=tBbPS4ZQU1BKmBEn7Svf3A&_nc_ss=7b20f&oh=00_AQMw7FRu2qfIB1kPYHFoJUHTKQ-RuexbeycAZxgTIkMEsw&oe=6AEFB63E", "summary": "Lawrence Berkeley National Laboratory — one of the US Department of Energy's premier research laboratories, known for Nobel Prize-winning work in physics, chemistry, and materials science — operates some of the most advanced scientific…", "tags": [ "ai" ], "title": "How Meta’s AI Models Are Powering the First Wave of Genesis Mission Projects", "url": "https://aggr.aymericbeaumet.com/items/ai-meta-com/2026-07-21-how-meta-s-ai-models-are-powering-the-first-wave-of-genesis/" }, { "content_html": "Today, we’re excited to introduce Muse Spark 1.1, the latest model from Meta Superintelligence Labs and a significant upgrade from Muse Spark. Muse Spark 1.1 is a multimodal reasoning model built for agentic tasks, with major gains in tool and computer use, coding, and multimodal understanding.
\nWith these improvements, Muse Spark 1.1 advances the performance-efficiency frontier. Together with this week’s launch of Muse Image, this release brings us closer to our vision of personal superintelligence: models that help you pursue your goals, create what you imagine, deepen your relationships, and take action on what you value most.
\nAlong with this release, we are launching a public preview of the new Meta Model API where developers can access Muse Spark 1.1. The model is available now in "Thinking" mode in the Meta AI app and on meta.ai.
\nEvaluations
\nAgents
\nMuse Spark 1.1 delivers exceptional performance in personal agentic tasks that require planning and orchestration across a range of external apps and services. It zero-shot generalizes to new native tools, MCP servers, and custom skills.
\nIt tackles complex projects significantly faster than Muse Spark, as it is trained to orchestrate multi-agent systems to optimize end-to-end latency. As the main agent, it can gather context, make a plan, and delegate execution across parallel subagents. As a subagent, it adheres to its job, understands available tools, and knows when to escalate back to the main agent.
\nMuse Spark 1.1 can actively manage its context window of 1 million tokens. It remembers actions, retrieves information from much earlier work, and compacts in a way that keeps the critical steps needed for later work.
\nComputer Use
\nMuse Spark 1.1 excels at computer-use workflows that unfold across multiple applications with information changing on-the-fly. It maintains context across extended sessions, adapts to evolving requirements, and navigates unfamiliar interfaces with minimal human intervention.
\nRather than reasoning through every desktop step one click at a time, Muse Spark 1.1 understands when to automate and when to use the interface directly. We trained the model to write scripts when automation is faster, click when direct interaction is simpler, and generate batches of actions at each step.
\nAgentic dinner party organization: In real-world applications, new context arises that changes the task. Muse Spark 1.1 notices these changes when placing the dinner order and makes necessary updates without user intervention.
\nCoding
\nCoding performance for Muse Spark 1.1 improved substantially on real-world tasks involving large, complex codebases. It can diagnose and fix complex bugs, implement new features in enterprise-grade systems, and execute large code migrations. In use cases like creating web applications and end-to-end question answering, Muse Spark 1.1 shows large gains over our first model.
\nWe trained our model to smoothly adapt to diverse harnesses and reliably handle complex multi-turn dynamics. Muse Spark 1.1 performs well with popular agentic coding setups, supporting common features like planning mode, goal conditioning, subagent delegation, and context compaction.
\nDebugging demo in OpenCode: Muse Spark 1.1 builds a chat web app, takes automated screenshots to identify user-visible failures, traces issues back to relevant code to implement fixes, and validates these changes. The model seamlessly combines coding, multimodal understanding, and tool calling.
\nAcross Meta, developers and researchers are using Muse Spark 1.1 daily to build faster and work smarter. On our primary internal coding evaluation, Meta Internal Coding Bench, Muse Spark 1.1 significantly improves upon Muse Spark and is competitive with leading alternatives.
\nResearchers are now also automating model development and evaluation tasks by leveraging Muse Spark 1.1 in their workflows.
\nDeepSWE evaluation in OpenCode: Muse Spark 1.1 evaluates itself on a subset of DeepSWE tasks across different reasoning strengths and produces an analysis dashboard based on the results.
\nMultimodal
\nAlong with coding and agentic capabilities, Muse Spark 1.1 excels in perception, multimodal reasoning, and tool use. It can interact with real environments and produce grounded outputs with strengths in visual-to-code artifact generation, ultra-descriptive image and video captioning, and agentic workflow execution for multimodal use cases.
\nMuse Spark 1.1’s multimodal capabilities are especially valuable when perception and action need to happen together. The model can inspect visual and audio, preserve details across a long workflow, and use those details while operating computers on the user’s behalf.
\nFacebook Marketplace agent: Using video shot from a smartphone, Muse Spark 1.1 extracts useful photos and reasons about the product to operate a user's browser and make a Facebook Marketplace listing on the user's behalf.
\nSafety
\nWe conducted extensive safety evaluations before deployment, following the Advanced AI Scaling Framework, which defines evaluations, threat models, and deployment thresholds for our most advanced models.
\nAcross all frontier risk categories — Chemical & Biological, Cybersecurity, and Loss of Control — our evaluations show Muse Spark 1.1 operates within safe margins. Muse Spark 1.1 demonstrates strong resistance to direct jailbreaks and indirect attacks from untrusted data, prompt injection, and developer-prompt attacks. Consequently, it shows better adversarial robustness, lower hallucination rates, and reduced sycophancy.
\nOur full safety posture for 1.1 is documented in our Muse Spark 1.1 Evaluation Report.
\nAvailability
\nFor the first time, developers can begin building with Muse Spark 1.1 via the new Meta Model API, now in public preview. Early partners of Muse Spark 1.1 praise the model as a complete agentic foundation, pairing long context handling with strong coding and reasoning capabilities to handle large-scale agentic workloads.
\n“What’s most impressive about Muse Spark is how much it packs into one model: massive million-token context, full multimodal support (images, video, PDFs), built-in search with citations, strong reasoning, top-tier coding abilities (particularly frontend and design), structured output, and parallel tool calling — all in a clean OpenAI-compatible package. A complete agentic foundation."
\n— Amjad Masad, CEO of Replit
\n“Meta is clearly building for serious agentic coding – strong tool use at a price point that makes it viable to run real coding workloads at scale. That combination is rare, and it’s exactly why we wanted Cline developers to have access early.”
\n— Saoud Rizwan, CEO of Cline
\n“When tested against Box’s enterprise work evaluation set, Muse Spark delivered enterprise capabilities competitive with today's leading frontier models. That level of intelligence, combined with its strengths in structured, procedural workflows across industries such as professional services, public sector, and industrial operations, makes it a compelling choice for organizations.”
\n— Yashodha Bhavnani, VP of AI Products at Box
\n"Muse Spark 1.1 is an awesome model for running agents. Fast, powerful, and fun with OpenClaw.”
\n— Dave Morin, OpenClaw Foundation
\nWe're thrilled to be releasing Muse Spark 1.1, a testament to our research momentum. We have even more capable models in training and look forward to sharing what’s to come.
\nWritten by:
\nMeta Superintelligence Labs
\nShare:
\n", "date_published": "2026-07-09T00:00:00+00:00", "external_url": "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/", "id": "urn:aggr:item:d4398d835bbc6efe1a2fd0dc642e947dc9b2e2de", "image": "https://scontent-sjc6-1.xx.fbcdn.net/v/t39.2365-6/741906678_873652302465180_7764547737953411008_n.png?_nc_cat=104&ccb=1-7&_nc_sid=e280be&_nc_ohc=wLMAAGlsRQEQ7kNvwH4SYhT&_nc_oc=AdqtCqt5J2zSGDJuoYCrIaj5Zm9U88KpKk_Rzq-jj7WtLvuqaG_IqCf3FCBJ9KYVyDo&_nc_zt=14&_nc_ht=scontent-sjc6-1.xx&_nc_gid=d8NecG7Z_F7rPYIpXoYtKg&_nc_ss=7b20f&oh=00_AQNJtPS-hbnyMYAoc6IW3o47gBZ_SjCg5ARFCYMLoPfljw&oe=6AEFB6E6", "summary": "Today, we’re excited to introduce Muse Spark 1.1, the latest model from Meta Superintelligence Labs and a significant upgrade from Muse Spark. Muse Spark 1.1 is a multimodal reasoning model built for agentic tasks, with major gains in tool…", "tags": [ "ai" ], "title": "Introducing Muse Spark 1.1", "url": "https://aggr.aymericbeaumet.com/items/ai-meta-com/2026-07-09-introducing-muse-spark-1-1/" }, { "content_html": "We’re excited to launch Muse Image and preview Muse Video, the first media generation models developed by Meta Superintelligence Labs.
\nMuse Image is our most advanced image generation model yet: it follows instructions faithfully, edits with precision, and composes from multiple references. It also brings agentic tool use capabilities and integrates with Muse Spark. Muse Video, built on the same pretraining base, delivers exceptional visual fidelity with native audio support.
\nMuse Image is available today across the Meta AI app and on meta.ai, Instagram Stories in the US, and WhatsApp in limited countries, and is coming soon to Facebook. Muse Video is coming soon to creators and Meta AI.
\nMuse Image: Agentic Image Generation
\nInstead of directly mapping prompts to images, Muse Image operates as an agent: it invokes search and coding tools to improve accuracy, self-refines its own generations, and improves through scaling test-time compute. Muse Image also integrates with Muse Spark, allowing the two models to share tools and plan jointly for powerful agentic media generation.
\nTool Use
\nWe provide Muse Image with access to tools to enhance its agentic capabilities.
\nCoding. During reinforcement learning, Muse Image learns to write and execute code that produces accurate plots and QR codes, and condition on rendered figures to improve the accuracy of generated images. Muse Spark and Muse Image also integrate to use the combination of code and media generation to create animated GIFs, websites with embedded images, and interactive visual games.
\nSearch. Muse Image learns to search the web to ground generated images in factual and real-time information and visual references. Enabling search improves factual accuracy on knowledge-intensive prompts, particularly those involving current events and real-world facts.
\nMuse Image improves with search tool use. Win rate from internal ablation.
\nSelf-Refinement
\nMuse Image reflects on and improves upon its own work within its chain of thought. This self-refining behavior can take different forms: a local edit to the current image draft when a small detail is off, a new image generation from scratch when larger parts are wrong, or a different tactic like tool use for more factually accurate generation. We didn’t design this behavior. Instead, it emerged during RL training simply because self-refinement produced better images and therefore higher reward.
\nMuse Image improves with emergent self-refinement. Win rate from internal ablation.
\nSelf-refinement
\nFinding reference images
\nSelf-refinement
\nAssembling the spread
\nI'm putting together the glossy magazine page, weaving the proof steps, headline, and fashion portrait into a cohesive layout while checking typography, spacing, and the gold-accented details for a polished editorial feel.
\nSelf-refinement
\nRefining the formula
\nI'm correcting the formula to include the missing division slash, updating the layout so the expression reads S = n(n + 1) / 2, and double-checking the surrounding text for clarity and consistency.
\nSelf-refinement
\nChecking the image
\nI'm reviewing the generated magazine image to confirm the macro detail, glossy layout, and accurate proof before sharing it, and I'm ready to tweak the typography or swap the portrait if you'd like. What would you like to adjust next?
\n✓
\nFinished
\nTest-Time Compute Scaling
\nLike language models, Muse Image improves the more it thinks at inference time. With more test-time compute, the model reasons more, uses more tool calls, and uses more self-refinement steps to improve its generations. Increasing reasoning strength (and thus test-time compute) improves human-preference Elo scores and shows an approximately log-linear scaling relationship. Notably, this compute spans two very different kinds of work — text tokens for reasoning, visual tokens for generation — yet quality is a function of the combined total compute.
\nWe find that using the token budget judiciously matters just as much for effective test-time scaling. Best-of-N (BoN), where the model generates several images and keeps the best, improves quality early but saturates quickly. Spending that same compute on deliberate reasoning scales considerably better. Reasoning and tool use compound when combined. Tools let the model reach beyond what it already knows, whether by searching for references it lacks or writing code to get precise details right, filling gaps that reasoning alone can’t.
\nMuse Image improves with scaling test-time compute. Elo from internal ablation.
\nImage Editing
\nMuse Image edits images with precision, changing exactly what the user asks for. It can follow a variety of instructions as our examples show.
\nMuse Image maintains coherence across editing turns, supporting iterative refinement and open-ended brainstorming toward a target result.
\nMulti-Reference Image Composition
\nMuse Image can compose elements from many input reference images in the prompt, including people, objects, clothing, styles, and environments. It supports interleaving text and images inline in prompts for complex image compositions.
\nImage Benchmarks
\nMuse Image holds the No. 2 spot on Arena for text-to-image, single-image editing, and multi-image editing as measured by human preference Elo rankings at the time of writing.
\nArena Elo rankings as of July 5, 2026.
\nArena Elo rankings as of July 5, 2026.
\nArena Elo rankings as of July 5, 2026.
\nPreviewing Muse Video
\nAlongside the release of Muse Image, we’re sharing an early preview of Muse Video. It offers competitive performance in prompt adherence, visual fidelity, and temporal consistency. We’re investing in areas with current performance gaps, such as audio-video synchronization and physically accurate fast motion. Muse Video is coming soon to creators and in Meta AI.
\nOn Arena, Muse Video ranks No. 3 in human-preference Elo for text-to-video at the time of writing.
\nArena Elo rankings as of July 5, 2026
\nContent Seal
\nTo help people verify whether an image is AI-generated, Muse Image includes Content Seal, our invisible watermarking system. Images created by Muse Image in the Meta AI app and on meta.ai carry a hidden provenance signal that stays intact — even when cropped, compressed, resized, or screenshotted. We plan to extend Content Seal to video soon. We’re previewing a detection tool that lets you check whether an image carries a Content Seal watermark, providing an initial way to help you better understand if an image was made with Meta AI.
\nMuse Image in Meta Products
\nMuse Image connects deeply with the Meta ecosystem. Our ongoing investments in image and video generation will further enable creators and businesses to generate dynamic content across Meta products.
\nMarketing assets for small businesses
\nPersonalized presets directly in Instagram
\nWritten by:
\nMeta Superintelligence Labs
\nShare:
\nOur latest updates delivered to your inbox
\nSubscribe to our newsletter to keep up with Meta AI news, events, research breakthroughs, and more.
\n", "date_published": "2026-07-07T00:00:00+00:00", "external_url": "https://ai.meta.com/blog/introducing-muse-image-muse-video-msl/", "id": "urn:aggr:item:50d41f8095d43a8e3de5b9e97aae48f14fe3e188", "image": "https://scontent-sjc3-1.xx.fbcdn.net/v/t39.2365-6/740903239_1399413145578607_7846806312558871713_n.webp?_nc_cat=110&ccb=1-7&_nc_sid=e280be&_nc_ohc=Z2GOMU_xLbAQ7kNvwE_ZFhC&_nc_oc=Ado-a2ZcpsKk5UzSRiKiVDCrwd9zu_xiIo1UtJu_i3C9eQ8e49V5foQagq8qVqTE3pM&_nc_zt=14&_nc_ht=scontent-sjc3-1.xx&_nc_gid=dhNBG5p7m_Deds03qBA_6Q&_nc_ss=7b20f&oh=00_AQOXeBSX48z77YumdIR5MrmXiQyxA6ZUD7SlUH46eqoK9w&oe=6AEFA3E1", "summary": "We’re excited to launch Muse Image and preview Muse Video, the first media generation models developed by Meta Superintelligence Labs. Muse Image is our most advanced image generation model yet: it follows instructions faithfully, edits…", "tags": [ "ai" ], "title": "Introducing Muse Image and Muse Video", "url": "https://aggr.aymericbeaumet.com/items/ai-meta-com/2026-07-07-introducing-muse-image-and-muse-video/" }, { "content_html": "Last year, we introduced Brain2Qwerty v1, research that uses AI to decode brain activity into text without any surgical implant. Now we're sharing the next step: Brain2Qwerty v2, the highest-performing end-to-end pipeline capable of real-time sentence decoding from non-invasive brain recordings, approaching levels of accuracy previously exclusive to techniques that require brain surgery.
\nTo help accelerate neuroscience breakthroughs, we're releasing the full training code for Brain2Qwerty v1 and v2, and our partner, the Basque Center on Cognition, Brain, and Language (BCBL), is releasing the v1 dataset. We believe this research has the potential to make a real difference for the millions of people who suffer from brain lesions that prevent them from communicating. Invasive procedures like stereotactic electroencephalography and electrocorticography have shown that a neuroprosthesis feeding signals to an AI decoder can restore communication, but they're difficult to scale. Our non-invasive approach can help bridge that gap.
\n![\"\"](\"https://scontent-sjc6-1.xx.fbcdn.net/v/t39.8562-6/731199808_1542868887492186_7676795253949750980_n.png?_nc_cat=101&ccb=1-7&_nc_sid=f537c7&_nc_ohc=H9r8BD-9XyIQ7kNvwES1gLy&_nc_oc=AdplBdNYmMloSZE-qvfFygg5fr6Ccn_-ozH8Uo4fHL9vUoC_AaYD8XzajqvCs_-vUn8&_nc_zt=14&_nc_ht=scontent-sjc6-1.xx&_nc_gid=vPdbxWJzwj5XTWluIdBYWA&_nc_ss=7b20f&oh=00_AQOdSWrEH8C3qLpIODvKIyujLADl-gcBCf8h7_zHsMIrkQ&oe=6ADB36DC\")
We trained Brain2Qwerty v2 on approximately 22,000 sentences from nine volunteer participants, each recorded for 10 hours wearing a magnetoencephalography (MEG) device while actively typing. Instead of relying on hand-crafted pipelines to detect neural events, we use end-to-end deep learning to decode directly from raw brain signals.
\n![\"\"](\"https://scontent-sjc3-1.xx.fbcdn.net/v/t39.2365-6/731219869_1076086811651166_7777400260549956173_n.png?_nc_cat=103&ccb=1-7&_nc_sid=e280be&_nc_ohc=DZzwpmRDJywQ7kNvwFt2Pdd&_nc_oc=Adq-YSC-JU0Bkn56KCdjkQI26kJ_pHc928ZQPPDjkhg9FFa3e_ip6RljWmz61Svm7Zo&_nc_zt=14&_nc_ht=scontent-sjc3-1.xx&_nc_gid=vPdbxWJzwj5XTWluIdBYWA&_nc_ss=7b20f&oh=00_AQMBjZcWHrozp0oLkTIbuMSyLfLKHWreU5tXP44qv6GmUw&oe=6AEF9E33\")
Fine-tuning large language models on neural data allows the system to leverage semantic context, bridging the gap between noisy brain recordings and coherent language. We also deployed AI agents to explore optimizations for the decoding pipeline, with final training configurations selected manually by engineers.
\n![\"\"](\"https://scontent-sjc6-1.xx.fbcdn.net/v/t39.2365-6/731158975_1696243454915298_6767206741376253564_n.png?_nc_cat=100&ccb=1-7&_nc_sid=e280be&_nc_ohc=yEolp2gO8YAQ7kNvwGqQLfp&_nc_oc=AdrJFyT6VaE4vtl7ebTJdRNaWd_bblt-irooxpB-xOgiTvAAT_AnCdoKToEmIHDjlkk&_nc_zt=14&_nc_ht=scontent-sjc6-1.xx&_nc_gid=vPdbxWJzwj5XTWluIdBYWA&_nc_ss=7b20f&oh=00_AQOt6l9KXWhVYNVpSeIrMQIavtpQb3iiTAs-bL3l8DF6Pg&oe=6AEFB0B3\")
The result: Brain2Qwerty v2 recovers sentences coherently from noisy neural inputs, achieving a word accuracy rate of 61%, significantly improving upon the 8% word accuracy from other non-invasive methods. And for our best participant, we achieve a 78% word accuracy, where more than half of all sentences are decoded with one word error or less.
\nWe also find that decoding accuracy improves log-linearly with data volume, suggesting that the remaining performance gap with surgical approaches could be further narrowed through data scaling alone. This work contributes to our efforts to build open foundational models of the brain, with our Tribev2 model for perception encoding, NeuralSet to process brain data at scale, and NeuralBench to systematically evaluate models. We do this in close collaboration with the community, through our recent $5 million fund to stimulate open datasets in our Digital Brain Project. Our hope is that this work, done in the open, advances neuroscience to identify, diagnose, and treat neurological disorders faster than in siloes.
\n", "date_published": "2026-06-29T00:00:00+00:00", "external_url": "https://ai.meta.com/blog/brain2qwerty-brain-ai-human-communication/", "id": "urn:aggr:item:66fa3d6b28cac1483a8e8964e24ba0f0c0f4d75b", "image": "https://scontent-sjc6-1.xx.fbcdn.net/v/t39.2365-6/731147471_987177500749005_1892166673921917145_n.gif?_nc_cat=101&ccb=1-7&_nc_sid=e280be&_nc_ohc=I6lZV37XGj8Q7kNvwGOQRri&_nc_oc=AdoId63SXpSCX-nqQ2oE48d8KkNB9XOpbTUmYKtctQ03OopBBPCSwaFU0xakwp5A9RI&_nc_zt=14&_nc_ht=scontent-sjc6-1.xx&_nc_gid=vPdbxWJzwj5XTWluIdBYWA&_nc_ss=7b20f&oh=00_AQOXZhQ1aWYO6iVeJG54PQgJtUnRE5S89so3w6KCRPoEGQ&oe=6AEFC9ED", "summary": "Last year, we introduced Brain2Qwerty v1, research that uses AI to decode brain activity into text without any surgical implant. Now we're sharing the next step: Brain2Qwerty v2, the highest-performing end-to-end pipeline capable of…", "tags": [ "ai" ], "title": "From Brain Waves to Words: Brain2Qwerty Offers a New Path to Communication Without Surgery", "url": "https://aggr.aymericbeaumet.com/items/ai-meta-com/2026-06-29-from-brain-waves-to-words-brain2qwerty-offers-a-new-path-to/" } ], "language": "en", "title": "AI at Meta Blog", "version": "https://jsonfeed.org/version/1.1" }![\"Text-to-image](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=1566388491585031&version=1789412025\")
![\"Single-image](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=1066985612654620&version=1789412025\")
![\"Multi-image](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=1793757385125329&version=1789412025\")
![\"Inference-time](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=1287728376563752&version=1789412025\")
![\"Inference-time](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=1031626362749236&version=1783971604\")
![\"Inference-time](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=36940809352231111&version=1783971604\")
![\"Inference-time](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=1472584831306392&version=1783971604\")
![\"Inference-time](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=1714368106563655&version=1783971604\")
![\"Search-enabled](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=1020234107374296&version=1783727678\")
![\"Self-refinement](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=1051639497205607&version=1783727678\")
![\"Self-refinement](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=1278631507464493&version=1783727678\")
![\"Inference-time](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=1877978896921254&version=1783727678\")
![\"Text-to-image](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=1329479142728807&version=1783727678\")
![\"Single-image](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=1709492773435242&version=1783727678\")
![\"Multi-image](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=3481998118643042&version=1783727678\")
![\"Muse](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=1577457450453885&version=1783727678\")
![\"Avery](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=2209464969787484&version=1783727678\")
![\"Muse](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=1007664215499569&version=1783727678\")
![\"Muse](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=2466267673879898&version=1783727678\")
![\"Muse](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=2811407102591921&version=1783727678\")
![\"Muse](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=3315109218694153&version=1783727678\")
![\"Muse](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=996742523227072&version=1783727678\")
![\"Muse](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=1561261712375095&version=1783727678\")
![\"Muse](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=2510049522741124&version=1783727678\")
![\"Muse](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=1041312411770257&version=1783727678\")
![\"Muse](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=1330703112600334&version=1783727678\")
![\"Muse](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=1500373454711814&version=1783727678\")
![\"Muse](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=1661874731778824&version=1783727678\")
![\"Muse](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=1539507954301171&version=1783727678\")
![\"Muse](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=2525997914511953&version=1783727678\")
![\"Muse](\"https://lookaside.fbsbx.com/elementpath/media/?media_id=1018569877833951&version=1783727678\")