What if a machine vision system could be tested before a single component is purchased? What if thousands of defect images could be generated without a single physical sample? And what if setting up a virtual camera system were as simple as a conversation? Medabsy is addressing exactly these questions with two major additions to its platform: the Pro Version and Vision Navigator.
From virtual previews to production-ready datasets
The Light Version creates preview images for evaluating a machine vision setup. The Pro Version goes further and generates complete synthetic training datasets of photorealistic images, hundreds to tens of thousands of them. These images are not AI-generated: they are computed from a mathematical model of scene, material, defects, camera, optics and illumination, so no real dataset is needed to create one. This fundamentally changes the AI inspection workflow. Real-world training data is one of the biggest bottlenecks in AI project: engineers wait for production samples, deliberately damage parts or hunt for rare failure cases. Some defects are simply too rare to collect. Simulation offers a way out.
Defects on demand
Defect simulation is the Pro Version’s standout feature. Defects are placed directly on the virtual parts, with full control over size, shape, variation and position. Engineers can generate the scenarios their AI model actually needs, not the ones the production line happens to deliver. It is up to the users to decide what they want: rare defect types, extreme contrast, poor lighting, unusual part positions or several defects in one image. All created deliberately and in any quantity. Each defect image comes with pixel-perfect labels, removing most manual annotation work and introducing label consistency. The result is a controlled pipeline from virtual scene to synthetic training dataset.
Realism matters
Synthetic data is only useful if it reflects reality and can be trusted. Here, realism comes from physics, not generative AI: every image is rendered from mathematical models of light, optics and sensor behavior, not inferred from existing photographs. Generative models cannot be fully constrained or generalize beyond what they have seen, so their output may contain artefacts (hallucinations) that never occur in the real scene. A physically simulated image contains only what was defined in the setup, making the dataset deterministic, reproducible and auditable. This is where the two versions differ: the Light Version delivers preview images for first setup evaluation such as camera field of view and light coverage, the Pro Version delivers images realistic enough for evaluating defect visibility and performing AI training. With it, when designing a system, it is possible to move from „Will this camera system work?“ to „How will it perform across thousands of realistic scenarios?“ But between simulation and a real system lies one more hurdle: documentation. The Pro Version generates it automatically: bills of materials, technical drawings and data sheets, straight from the virtual design. That way the virtual model becomes the starting point for engineering and physical integration.
Fewer iterations, faster development
For system integrators, synthetic training data accelerates AI projects and makes them more reliable, especially where real data is scarce. Projects move faster because cameras, optics, lighting and inspection scenarios are evaluated before anything is built. Edge cases – the rare scenarios that typically cause a system to fail in the field – are validated up front rather than discovered in production.
Vision Setups through conversation
If the Pro Version expands what users can achieve, Vision Navigator focuses on how easily they achieve it. The new AI assistant simplifies the first simulation of a camera system, no more configuring every parameter by hand. Describe what needs to be inspected, and the AI proposes a suitable virtual setup. Fine-tuning then follows with proven parameter settings. The approach ´Develop Machine Vision through conversation´ may be the most important development of all: simulation becomes accessible to anyone who needs to know whether an inspection concept can work. It is particularly valuable for non-specialists: instead of learning the whole simulation environment first, they ask questions directly while the chatbot translates requirements into a workable simulation. At the same time, experienced engineers spend less time on setup; newcomers face a much lower barrier to entry.
Summary
The Pro Version and Vision Navigator address two sides of the same challenge. The Pro Version provides depth: physically simulated images, synthetic datasets without real training data, defect simulation, labels and automatic documentation. Vision Navigator provides accessibility: an AI-driven interface that gets users from idea to virtual camera system faster. The result is a different development model: explore, test and optimize virtually rather than build hardware first and discover its limits later. Synthetic data stops being an afterthought once real data turns out to be missing; it is there from day one. The machine vision system of tomorrow may not begin on the production floor. It may begin as a conversation – and be tested thousands of times before it ever becomes physical.
