Meta's AI models are being used to power the first wave of Genesis Mission projects, which aim to advance scientific research and exploration. This collaboration leverages Meta's advanced AI capabilities to support innovative projects in the Genesis Mission initiative.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
The U.S. Department of Energy's Lawrence Berkeley National Laboratory and other institutions use Meta's open-source models SAM 3 and DINOv3 to reduce scientific image segmentation time from weeks to 15 minutes, paving the way for real-time scientific discovery.
The DOE's synchrotron and neutron source facilities generate tens of petabytes of data annually, equivalent to streaming 2 million hours of HD video. After detector upgrades, capture rates increased from one image every 6 seconds to 100,000 images per second, with data volumes orders of magnitude higher than a decade ago. Traditional manual analysis cannot handle this scale, and domain experts are scarce. Modern in-situ experiments require real-time interpretation, which manual teams cannot provide.
In late 2025, the White House launched the Genesis mission, led by the DOE, to accelerate scientific discovery using AI. The SYNAPS-I project, part of this mission, focuses on transforming X-ray and neutron science data analysis from a months-long bottleneck into a real-time engine. Image segmentation is a key task: identifying and outlining structures such as cell walls and mineral grains from raw X-ray images.
SYNAPS-I employs two open-source foundation models from Meta: SAM 3 for precise pixel-level boundary segmentation and DINOv3 for global context to identify each structure. The team fine-tuned the models on scientific imaging data collected at DOE beamlines and deployed them on 300 A100 GPUs at the National Supercomputing Facility. Result: fully reconstructed, semantically annotated 3D volumes are returned to scientists while the experiment is still running, in about 15 minutes.
The SYNAPS-I team demonstrated the workflow at the Advanced Light Source (ALS) for an agricultural challenge: micro-CT scanning of grapevine stems to automatically identify xylem vessels responsible for water transport and track changes during drought. Previously, each time step required a month of expert annotation; now it takes 15 minutes, allowing scientists to study dynamic biological processes at the speed of data acquisition.
National labs must manage pre-publication research data and AI models on government infrastructure, not external cloud services. Meta's open-source approach enables the SYNAPS-I team to download, fine-tune, and deploy SAM and DINO in their secure computing environments, adapting models originally trained on natural images to the scientific domain.
The SYNAPS-I project involves 60 researchers from five national labs, aiming to make user facilities intelligent discovery platforms—where AI not only accelerates data processing but also helps generate hypotheses, recommend next experiments, and transfer knowledge across facilities. DOE Deputy Secretary Dario Gil stated at the Trillion Parameter Consortium meeting that SYNAPS-I compresses discovery time from days to moments, establishing continuously self-improving scientific models, which is crucial for realizing the potential of the Genesis mission.
This article's information primarily comes from the Meta AI official blog, a first-party source. All data, timelines, and statements are derived from that article, with no external knowledge introduced.
The application of Meta's open-source AI models SAM 3 and DINOv3 in the SYNAPS-I project demonstrates how AI can reduce scientific image analysis from weeks to minutes, laying the foundation for real-time scientific discovery and cross-facility knowledge sharing.
Primary report
Primary source