On March 28th, it was reported that Tencent AI Lab unveiled its self-developed 3D game scene automatic generation solution at the 2023 Game Developers Conference. Through AIGC technology, this solution helps developers create highly realistic and diverse virtual city scenes in a very short amount of time, greatly improving game development efficiency.
During the conference, Tencent demonstrated how to quickly build a 3D virtual city from scratch using AI. The built virtual city covers an area of 25 square kilometers and includes a road network of 130 kilometers, 4416 buildings and over 380,000 indoor mappings. In the past, constructing such a large-scale city often required multiple artists working for years to complete; however, with AI integration it only takes weeks.
According to the introduction, in the game scene, the core technology of AI-generated virtual cities mainly includes city layout generation, building appearance generation, and indoor mapping generation.
The effect of city layout generation mainly depends on the realism and diversity of road layout and building layout. Tencent‘s 3D virtual scene automatic generation solution helps AI understand the characteristics of roads and buildings in real cities by learning from satellite images, aerial photography, and other real data sources. This enables it to quickly generate realistic city layouts with rich visual details.
With the push of AI technology, the generation of urban layouts has truly achieved intelligence. Taking road layout as an example, using Tencent‘s 3D virtual scene to automatically generate solutions, developers only need to draw the main roads and coastlines of the city, and the AI model can fill in reasonable road network details within one minute. At the same time, developers can also flexibly modify parameters to quickly generate different variants or make adjustments and modifications to generated results until they obtain their ideal virtual city roads.
Experiments have shown that the entire process of generating road layouts can be completed in less than 30 minutes, which is hundreds of times more efficient than traditional methods. In addition, based on the generated road network, AI models can also quickly generate diverse building layouts to make virtual cities more varied in shape. This process generally takes less than 10 minutes.
In the process of generating building facades, it used to rely mainly on manual reference to photos and manually creating unique buildings. It would take 2-3 days to complete a single model. Therefore, only a small number of unique buildings were made in most games, resulting in limited diversity when randomly combined and making the generated scenes feel repetitive.
Tencent‘s 3D game scene automatic generation solution combines AI with procedural generation, speeding up the process of creating 3D buildings from 2D photos. The production time for a single unique building has been reduced to 17.5 minutes, making it possible to generate large numbers of diverse building facades at low cost and high efficiency.
At the same time, by using self-developed image generation capabilities, large numbers of uniquely designed 2D building photos can be generated in batches, greatly enhancing the richness of building facades.Indoor mapping refers to adding a perspective of observing the indoor environment on building windows, “mapping” out the indoor environment and increasing realism.
Games have always been an excellent scenario for training and validating AI capabilities. Tencent AI Lab has been continuously researching in the field of AI+ games for many years, exploring full-chain AI solutions covering game development to operate around two major directions: AI Bot and AIGC, helping the industry improve efficiency and upgrade experience. In the future, Tencent will also extend these innovative achievements and technical capabilities to more scenarios, accelerating the integration of AI with physical industries.