Moemate’s distributed character engine enabled concurrent execution of 128 virtual characters (with a memory footprint of <38MB), and its own neuro-render sharding technology reduced the GPU load to 17 percent of standard solutions. According to Nvidia’s 2023 benchmark test, when it renders 50 Moemate characters in parallel on an RTX 4090 graphics card, the frame rate was 120FPS and the video memory consumption was only 4.3GB, 2.7 times more efficient than Unity’s ECS architecture. After the debut of the technology in Tencent’s overseas version of “King of Glory”, available sparring roles were increased from 8 to 64, and players’ matching waiting time was shortened to 3.2 seconds (82% shorter than the original system).
In the commercial environment, Moemate’s API cluster was able to handle 2,400 role interaction requests (QPS) with latency below 21ms. The Netflix interactive show Black Mirror: Bandasnyky 2.0 used Moemate to deploy 32 branching story characters with character switch speed of 0.4 seconds upon user choice, doubling the story tree complexity to 2^18 paths, and doubling average audience replay rate from 12 percent to 67 percent. According to Newzoo, these multi-character systems can increase the LTV (life cycle value) of interactive content to $45.7/ user, 3.1 times more than linear narrative products.
Moemate’s cross-platform character migration protocol (99.98% accuracy) synchronizes behavior data of up to 12 characters on iOS, Android, and PC. Through this technology, the library of NPC characters is distributed across the service, thereby increasing the 23 million players worldwide that are engaging with some number of NPCS from 15 million to 230 million and reducing server expense by 29%. Its data compression algorithm compacts status for single roles to 0.7KB, or 41 percent lower bandwidth than Google Protocol Buffers.
To meet individual requirements, Moemate‘s genetic algorithm model could generate 100 character prototypes in 5.3 seconds on 72 cultural traits. Adobe Character Animator leverages this capability to increase the productivity of designer character creation to 42 per hour, 18 times faster than hand modeling. Pixar’s tests revealed that 87 percent of the 500 mass-generated cast characters with Moemate passed director artistic review (compared to the industry manual production rate of 34 percent).
In the area of multi-role cooperation, Moemate’s group behavior AI enabled autonomous cooperation among 256 virtual characters (98.5 percent conflict resolution rate). By using the technology on Nvidia’s Omniverse platform, the number of robot characters in the digital factory simulation scenario was increased from 80 to 1,200, and the usage of computing resources was reduced to 23% of the original system. Boston Dynamics Atlas robot choreographed 32 machines with the Moemate group control module successfully and reduced the error in motion synchronization from ±120ms to ±9ms.
From a security compliance point of view, Moemate’s role isolation system came with an ISO/IEC 27001 certification that contained a sandboxing feature in which there was a less than 0.003% chance of a data breach when 1,000 roles were run concurrently. With the deployment of this technology in Zoom’s latest virtual meeting platform, the number of supported roles in a meeting room jumped from 16 to 200, and CPU usage is once again steady at 18%-22%. Multi-role applications like this, according to Gartner, reduce enterprise virtual training costs to $0.17/ person · hour, 94% lower than the cost of traditional VR training.