NPCTalk, an open-source toolkit developed by Universal Modder, is adding dynamic real-time conversations to NPCs in Grand Theft Auto V. Players can speak directly through their microphone, and the system generates responses based on the character's personality, surroundings, and current context, then delivers them using cloned voices that closely mimic the original performances. The developer stated that a free-to-play version of Grand Theft Auto V is expected to launch next week.
The demo features characters such as Michael, Lamar, ordinary pedestrians, and police officers. Some NPCs not only respond to players but also accompany their dialogue with gestures, play animations, or execute specific verbal commands.
This technology stack consists of multiple models and services: NVIDIA Nemotron ASR handles speech recognition, Qwen 3.8 27B running on Cerebras generates dialogue, Fish Audio Drama 3 performs voice cloning and speech synthesis, and TypeSafe Jev manages NPC animations.
The developer initially claimed response times under one second, but technical documentation records actual push-to-talk latency at approximately 1.15 to 2.2 seconds, with open-mic detection adding an extra 0.7 seconds. Currently, the tool is only demonstrated in the single-player mode of Grand Theft Auto V Legacy Edition and is not a one-click mod applicable to all games; other titles require separate integration, and using cloud services and hardware like H100 GPUs may incur costs. The toolkit is now available on GitHub under the MIT license.