AI models used by Playworlds
AI models behind one continuous performance.
Playworlds gives each model family a focused job so the story can reason, listen, recognize a turn, and speak back as one coherent game experience.
Choose a worldCurrent production stack
Models with focused roles
The collection below reflects the production paths in this repository. Configurable experiments are excluded so the page describes the game players actually meet.
Layer 01 · Text to speech
MiniMax Speech 2.8 HD
Playworlds uses MiniMax text-to-speech for spoken game-master narration and AI companion dialogue. The default production configuration uses Speech 2.8 HD with separate voice identities for each role.
See its role in Playworlds
Layer 02 · Narrative reasoning and game state
Google Gemini family
Playworlds uses different Gemini models for different kinds of thought: deep game-master turns, quick in-character conversation, teammate actions, and selected translation tasks.
See its role in Playworlds
Layer 03 · Speech recognition
Sherpa-ONNX SenseVoice
Playworlds sends browser microphone audio to its voice gateway, where Silero detects speech and an int8 SenseVoice model running through Sherpa-ONNX transcribes supported languages without browser SpeechRecognition.
See its role in Playworlds
Layer 04 · Voice activity detection
Silero VAD
Playworlds uses Silero VAD as a supporting voice model. It helps distinguish likely player speech from silence and ambient sound so a spoken turn can begin, end, and recover naturally.
See its role in PlayworldsThe design principle
A model is only useful when the player stops noticing it.
Reasoning, listening, turn detection, and performance should feel like one uninterrupted exchange. Playworlds evaluates the stack by the rhythm of play—not by a leaderboard score in isolation.
Common questions
The short version
Does Playworlds use one AI model for everything?
No. It combines focused model families for narrative reasoning, speech recognition, voice activity detection, and text-to-speech.
Which Gemini models does Playworlds use?
The current production paths use Gemini 3.5 Flash for game-master and action reasoning, Gemini 3.1 Flash Lite for companion chat, and Gemini 2.5 Flash Lite for selected translation tasks.
Can I still play without voice input?
Yes. Text input remains available when you prefer typing or microphone access is unavailable.