Persona Studio

Persona Studio The AI Narrative Architect. Generate consistent worlds, simulate deep character psychology, and voice-chat in real-time. Future-ready with Google Genie 3 for instant playable prototypes

Comment

Inspiration

Every game developer and novelist knows the struggle of the "Dead Bible." You spend months writing a 100-page design document or world bible, but the moment production starts, it becomes obsolete. It’s static, searchable only by keywords, and disconnected from the creative process.

We asked: "What if your design document could talk back?"

We wanted to move beyond static text generation to active simulation. Instead of just writing a character's backstory, what if you could interview them? Instead of manually updating a timeline when you delete a faction, what if the history rewrote itself to maintain continuity? Inspired by the promise of Neural Game Engines, we built Persona Studio to be the first "living" architecture tool for narrative design.

What it does

Persona Studio is a privacy-first, local-first Narrative Architect that helps creators build, manage, and simulate complex story worlds.

  • World Genesis (The Spark): Users input a simple constraint (e.g., "Victorian Cyberpunk Mystery") or upload existing PDF lore. The system instantly generates a cohesive "World Bible," including a historical timeline, key factions, and a cast list—all grounded in that specific reality.
  • Deep Character Profiling: We go beyond RPG stats. The engine generates deep psychological profiles, defining a character's specific fears, cognitive biases, and speech quirks to ensure they feel human.
  • The Simulation Deck (Neural Link): The killer feature. Creators can open a real-time voice channel to "call" their characters. Using Gemini Live, you can interview your villain, test dialogue flow, or see how they react to specific game states—hearing their distinct AI-generated voice respond instantly.
  • Smart Asset Management: The "Magic Wand" tool refines prose on the fly, while our "Lore Sync" feature ensures continuity. If you delete a major faction, the AI automatically rewrites the world history to remove or retcon their involvement seamlessly.

How we built it

We built Persona Studio as a modern, lightweight web application using React 19, Vite, and Tailwind CSS.

The core intelligence is powered by the Google GenAI SDK (@google/genai), orchestrating a suite of specific models for different tasks:

  1. Gemini 3 Flash Preview: This is the "Prefrontal Cortex" of our app. We leverage its massive context window and superior reasoning for World Genesis and Lore Synchronization. It allows us to feed entire PDF reference documents into the context, ensuring the generated content adheres strictly to the user's uploaded lore.
  2. Gemini 2.5 Flash Native Audio: Powering the Simulation Deck, this model handles the low-latency, real-time voice interactions, providing distinct voices (Kore, Puck, Fenrir) for different character archetypes.
  3. Gemini 2.5 Flash Image: Generates high-fidelity concept art for characters based on their psychological profiles.
  4. Local Storage: To respect creator privacy, all data is persisted locally in the browser. No external databases required.

Challenges we ran into

  • "Generic Trope" Syndrome: Early versions of the World Genesis engine often defaulted to generic fantasy clichés. We solved this by building a "Chaos Engine"—a logic layer that forces random, conflicting seeds (Genre + Tone + Archetype) into the prompt to force the model out of its comfort zone (e.g., forcing it to combine "Regency Romance" with "Eldritch Horror").
  • Continuity Maintenance: Ensuring that a character's voice remained consistent between the text-based profile and the audio-based Simulation Deck was difficult. We solved this by serializing the entire psychological profile (fears, biases) into the system instruction of the Live API session.

Accomplishments that we're proud of

  • The "Alive" Factor: The first time we successfully "called" a generated character and heard them lie to us about their backstory (because their profile said they were "deceptive"), it felt like magic.
  • Seamless Multimodality: integrating text, audio, and image generation into a single, cohesive dashboard where they all inform each other. The image generation knows the character's backstory; the voice knows the character's fears.

What we learned

  • Context is King: The Gemini 3 context window is a game-changer for narrative tools. Being able to dump raw PDFs of research material into the prompt allows the AI to act as a true subject matter expert, not just a creative writing assistant.
  • Audio Latency: For "simulation" to feel real, latency must be near-zero. The Native Audio capabilities of Gemini 2.5 are essential for breaking the "uncanny valley" of conversational pacing.

What's next for Persona Studio

  • Integration with Genie 3: This is our north star. We plan to connect our World Genesis output directly to Google's Genie 3 World Model. This would allow users to not just read about the world they generated, but step into it—generating playable, logic-free prototypes of their levels and environments instantly.
  • Multi-Agent Simulation: Creating "Chat Rooms" where two AI characters can converse with each other while the writer observes.

Built With

Share this project:

Updates