The Roboflow Vision AI platform lets manufacturers stand up production-grade defect-detection models in days with the NVIDIA Defect Image Generation skill.
TAIPEI, June 1, 2026 —Roboflow today unveiled a breakthrough in training production grade defect detection models with limited data. The Roboflow platform now enables manufacturers to generate synthetic data with NVIDIA Defect Image Generation skill for defect image generation, Cosmos world foundation model, and NVIDIA Isaac Sim to close the data gap that has stalled visual inspection programs on plant floors for the better part of a decade.
In a benchmark run with Corning Incorporated's optical fiber manufacturing engineering team, a model trained on just eight real defect images plus synthetic examples generated by Cosmos reached 0.95 mean average precision and perfect recall on the toughest defect class, beating a baseline trained on real data alone.
In Corning’s buffering operations, the proof of concept demonstrated the potential to reduce daily manual image review, improve turnaround time for reject disposition, and accelerate model deployment from a line-by-line effort to a more scalable rollout approach, with potential to accelerate model deployment across vision-capable systems within Corning.
The defects that drive recalls, scrap, and customer escapes are, by definition, rare. Waiting for enough real examples to train a reliable inspection model can take a full year of production, and most inspection pilots never make it to the line for exactly that reason.
Integrated with the NVIDIA Defect Image Generation skill, the Roboflow synthetic data pipeline closes that gap by generating photorealistic, physics-consistent defect images grounded in a manufacturer's own product geometry and a small set of real reference examples, with controllable variation across lighting, camera position, and material conditions. The Roboflow Agent handles labeling, training, evaluation, and deployment to the cameras already on the line.
For manufacturers, that compresses a multi-quarter inspection project into a few days and offers an agentic solution to automate the entire workflow end-to-end compared to more rigid model training methods. It also removes the most common reason new product introductions stall on the line, which is the lack of defect training data.
The new Roboflow SDG pipeline extends to every new product line, every new geometry, every new defect mode a manufacturer will encounter. SDG is also accessible through the Roboflow agent and acts as a specialized visual inspection agent that a factory manager agent built with the NVIDIA Factory Operations Blueprint (FOX) can connect with to manage the end-to-end vision AI model training pipeline. Roboflow agent capabilities include data management, generation, labeling, training, evaluation, and deployment to the cameras on the line.
The lifecycle of a visual inspection agent stops being a one-time, manual integration project and starts behaving like an autonomous unified layer of agentic intelligence across simulation, training and deployment. The system improves the longer the line runs, and sits alongside existing inspection systems as a software layer that adapts without new capital expenditure
“For 175 years, Corning has paired materials science innovation with advanced manufacturing. The Roboflow Agent powered by NVIDIA allows us to generate the training data we need, fine-tune our models, and strengthen model performance and inspection quality while increasing the speed, scalability, and adoption of next-generation technologies,” said Jeremy Knopf, chief information officer, Corning Optical Communications.
"Roboflow exists to make the world programmable and a reliable synthetic data pipeline is the missing breakthrough for ubiquitous vision AI adoption in manufacturing. Every camera in every facility on the planet now becomes a programmable endpoint to deliver agentic AI solutions. NVIDIA, Corning, and Roboflow are showing the first real version of that future this week." said Joseph Nelson, CEO, Roboflow
The NVIDIA Defect Image Generation skill is openly available on GitHub, and implemented end-to-end on the Roboflow platform. Manufacturers can deploy the published workflow themselves or contact Roboflow to scope an on-site engagement.
About Roboflow
Roboflow gives organizations the platform to build visual intelligence that understands and acts in the physical world. The Roboflow platform spans labeling, training, deployment, and applications, and is used by more than one million engineers and over half of the Fortune 100 across manufacturing, logistics, rail, defense, healthcare, retail, and sport. Learn more at roboflow.com.
About Corning Incorporated
Corning (www.corning.com) is one of the world’s leading innovators in materials science, with a 175-year track record of life-changing inventions. Corning applies its unparalleled expertise in glass science, ceramic science, and optical physics, along with its deep manufacturing and engineering capabilities to develop category-defining products that transform industries and enhance people’s lives. Corning succeeds through sustained investment in RD&E, a unique combination of material and process innovation, and deep, trust-based relationships with customers who are global leaders in their industries. Corning’s capabilities are versatile and synergistic, which allows the company to evolve to meet changing market needs, while also helping its customers capture new opportunities in dynamic industries. Today, Corning’s markets include optical communications, mobile consumer electronics, display, automotive, solar, semiconductors, and life sciences.
Cite this Post
Use the following entry to cite this post in your research:
Trevor Lynn. (May 31, 2026). Roboflow Advances Defect Detection Pipeline for Manufacturers with Synthetic Data Powered by NVIDIA Physical AI. Roboflow Blog: https://blog.roboflow.com/synthetic-data-generation-manufacturing-nvidia/