[Edaily Marketin Song Seung-Hyeon Reporter] Every time the camera industry changes a single combination of lenses, sensors, and image processing, it creates samples, takes test shots, and then redesigns the system. As the number of combinations to review increases, so do development time and costs. QBlack AI has shifted this process to software. By running the six stages through which a camera perceives the world—from the light source to the AI algorithm—within a single pipeline, it identifies which combinations are viable before creating prototypes.

According to the venture capital (VC) industry on the 14th, QBlack AI secured seed funding from Partners Lounge and Korea Investment Accelerator. The investment amount and company valuation were not disclosed. The company plans to use the funds to hire development personnel and advance its simulation technology. The company was also selected for “Barun Donghaeng,” an early-stage investment program run by Korea Investment Accelerator, as well as the “US Forged” program operated by D.Camp and HAX.

The company was established on February 11. CEO Kim Jun-sung is an engineer with experience in camera module mass production. In the company’s introductory materials, CEO Kim wrote that lenses, sensors, image processing, and algorithms must be designed together from the very beginning. Explaining the background of the startup, he said, “I personally experienced the limitations of these stages being developed separately on the camera module mass production floor,” adding, “The process of creating a sample, taking photos, and redesigning it to verify a single combination was repeated over and over.”

The name of the platform currently under development is “QBLACK.” It physically integrates six components—light source, scene, lens, sensor, ISP (image signal processor), and AI algorithms—based on physical principles. Users input specifications such as the curvature, thickness, and material of each lens surface; the pixel size, quantum efficiency, and noise level of the sensor; and the white balance and defective pixel correction values of the ISP. The system then displays the final image that would likely result from that combination, along with performance metrics for object detection, depth estimation, and image segmentation—all on a single screen. It takes 1–2 minutes to run a single combination on a graphics processing unit (GPU) and about 10 minutes on a central processing unit (CPU).

What’s noteworthy is the approach to handling synthetic data. Existing synthetic images created with game engines—no matter how realistic they may be—do not pass through the lenses and sensors that would be used in an actual product. This is why there is a discrepancy between the images the AI has been trained on and the footage captured in the field. QBlack AI bridges this gap by converting synthetic images into results that have passed through the target camera’s optics, sensor, and image signal processor (ISP). The company describes its technology as “redefining computer vision in terms of photons rather than pixels.”

The revenue model is subscription-based. However, the company does not sell the technology outright. It offers a free trial through website registration and first undergoes a joint validation process with interested companies. The process consists of four stages—consultation, simulation, actual measurement, and validation—to confirm that the simulation results match actual measurements, after which a subscription conversion is discussed. During this period, the software is provided free of charge, and the test hardware is co-developed by both companies. Companies participating in the validation receive one year’s worth of service free of charge.

The company targets markets such as mobile imaging, autonomous driving and Advanced Driver Assistance Systems (ADAS), robotics and humanoids, industrial machine vision, and AR/VR—areas where the end-user of the camera is an algorithm rather than a person.

Among companies that have taken a similar approach is Spain’s Aniversus, which specializes in sensor simulation for autonomous driving. Tools in the NVIDIA Omniverse suite also handle scene-level synthetic data. QBlack AI emphasizes that it addresses the internal physics of the camera at the design stage. Grand View Research projects that the synthetic data generation market will grow at a compound annual growth rate (CAGR) of 35.3% through 2030. The company plans to begin full-scale technology validation with U.S. firms.