Inspiration
We grew up playing "Choose Your Own Adventure" books and RPGs, but we were always limited by the pre-written paths and static descriptions. I wanted to build something that felt like a living, breathing cinematic universe—one where the world doesn't just respond to your choices but visualizes them in real-time. With the release of Gemini 3, we realized we finally had the "Director" capable of orchestrating a truly infinite, high-fidelity narrative.
What it does
Chronicles is an AI-powered interactive cinematic engine. Users start by defining a theme—like horror, murder mystery, action.
Dynamic Narrative: Gemini 3 acts as the Dungeon Master, generating complex branching storylines with deep character arcs.
Visual Immersion: Nano Banana Pro generates high-fidelity keyframes and character portraits that maintain visual consistency across scenes.
Cinematic Motion: Veo 3 transforms pivotal story moments into breathtaking 4K video clips with synchronized audio, making the user feel like they are watching a movie they are starring in.
Player Agency: At every turn, the user is presented with choices that fundamentally alter the world state, tracked precisely by Gemini’s massive context window.
How I built it
I leveraged the Google AI Studio ecosystem to create a seamless multimodal pipeline:
The Brain (Gemini 3): I utilized Gemini 3 to manage state and generate narrative. It processes the user’s history and outputs structured JSON containing the next story beat, three branching options, and specific prompts for the visual models.
The Eyes (Nano Banana Pro): To ensure the art style remained consistent, I fed specific style seeds and character descriptions generated by Gemini into Nano Banana Pro for the interface's environmental art.
The Soul (Veo 3): For "Hero Moments," I triggered Veo 3 to generate 5-10 second cinematic sequences, using Gemini 3 to write the prompt-based "screenplay" and audio cues for the video.
The Voice (Gemini 2.5 Flash TTS): For the voice-over narration to maintain the tension.
Frontend: A React-based interface that feels like a premium streaming platform, designed to showcase the generated media without distraction.
Challenges we ran into
Visual Consistency: Keeping a character looking the same across different scenes is notoriously difficult. I solved this by having Gemini 3 generate "Character Style Sheets"—detailed physical descriptions that were appended to every Nano Banana and Veo prompt.
Latent Orchestration: Coordinating three massive models to respond in a timeframe that feels "interactive" was a hurdle. I implemented a predictive loading system where Gemini 3 "pre-calculates" potential story paths to start the image/video generation before the user even clicks their choice.
Prompt Leakage: Ensuring the AI didn't break the fourth wall or generate nonsensical options required rigorous system prompting and iterative testing in AI Studio.
Accomplishments that we're proud of
Zero-Shot World Building: I'm amazed at how Gemini 3 could take a simple theme and build a coherent storyline for that world instantly.
Cinematic Quality: The integration of Veo 3 creates moments of genuine "wow" factor that go beyond a standard text adventure.
Technical Synergy: Successfully chaining three different state-of-the-art models into a single, cohesive user experience.
What I learned
I learned that the future of storytelling isn't just about "generating content," but about orchestration. The magic happens in the handoff—how the LLM describes a scene to the video model determines whether the immersion holds or breaks. I also gained a deep appreciation for the Gemini 3 context window, which allowed the story to remember a small detail from "Scene 1" and bring it back for a payoff in "Scene 8."
What's next for Chronicles
Multiplayer Journeys: I want to allow groups of friends to vote on choices or play as different characters in the same generated world.
Voice Integration: Implementing Gemini Live's capabilities so users can speak to the characters and influence the story through natural conversation.
Persistent Universes: Allowing users to "publish" their generated worlds so others can play through the same timeline and see if they reach a different ending.
Log in or sign up for Devpost to join the conversation.