AI Verse offers a self-service image factory that produces high-quality annotated synthetic datasets for computer vision engineers. An entirely novel process enables the user to describe their ideal dataset and launch its generation on AI Verse’s render farm, a scalable cloud-hosted cluster of GPU machines, each able to procedurally build a 3D scene and render photorealistic images in a few seconds. The dataset builder offers simple but powerful inputs for scene description, lighting, and camera placement, thus emulating the work of a 3D artist in just a few clicks. How it works The user specifies the environment (e.g., living room, bedroom, office) and then selects object classes of interest from a catalog of 1,000 plus assets in an ongoing expansion. These inputs are enough to generate any desired number of 3D scenes procedurally, respecting user-defined constraints while offering the variations in appearance and content necessary for AI robustness on tasks such as object detection and semantic segmentation. The user can also specify activities to be performed by human agents in the scenes, ranging from simple postures to leisure or work-related activities (e.g., typing on a keyboard or watching TV). A wide range of lighting scenarios can be applied to simulate varying weather and time of the day. The camera placement is also randomized according to simple and effective constraints that can be defined according to the engineer’s use case. Sensor parameters (e.g., lens parameters, depth-of-field, exposure) can also be adjusted. The process of setting and adjusting scene and image parameters is made interactive and engaging through the use of live previews, allowing the user to visualize within a few seconds an image rendered on the fly on one of our GPU machines, corresponding to the input parameters. The availability of GPU machines for previews is guaranteed through a session-booking system.