Open-Source Mod NPCTalk Uses AI to Give GTA 5 NPCs Real-Time Unscripted Conversation

Universal Modder has released NPCTalk, an open-source AI modding toolkit on GitHub that enables real-time, unscripted voice conversations with non-playable characters (NPCs) in Grand Theft Auto V. Using a combination of AI models, the mod lets players speak to almost any NPC through a microphone, with characters responding in voice-cloned audio and reacting with animations or specific voice commands. According to the developer, a free-to-play GTA 5 mod based on the toolkit is scheduled for release in the week starting October 26, 2026.

Unlike traditional scripted dialogue systems, NPCTalk dynamically generates responses using several models: NVIDIA's Nemotron ASR handles speech recognition, Qwen 3.8 27B running on Cerebras generates dialogue text in real time, Fish Audio Drama 3 handles voice cloning and synthesis, and TypeSafe Jev controls character animations. The speech models run on cloud-based NVIDIA H100 GPUs.

The system's latency is reported at about 1.15 to 2.2 seconds in push-to-talk mode, with an additional 0.7 seconds when the microphone is left in continuous listening mode. The developer also used Claude Code during development and packaged the technical experience into a reusable agent skill set compatible with Claude Code, Codex, Cursor, Gemini CLI, and other code assistants. Each game requires separate adaptation, and the current demo runs only in the legacy single-player mode of GTA 5. The toolkit is open-sourced under the MIT license, though cloud services and GPU infrastructure required to run it will incur costs.