DreamScape: Interactive Story Worlds — Devpost Summary

Inspiration

We grew up on choose-your-own-adventure books, interactive fiction, and games like "My Story" where every choice changed the narrative. But those experiences were always flat — text on a page, or 2D screens. We kept asking: what if you could actually walk through the story? What if every decision didn't just change the plot, but generated an entirely new world around you?

When we saw what World Labs could do — generating navigable 3D environments from a single text prompt in under a minute — the idea clicked. What if we built a platform where anyone could create branching interactive stories, and every choice literally builds a new AI-generated world? Not just for one story, but as a creative tool that lets anyone become a storyteller — and lets players remix those stories into their own.

DreamScape was born from the intersection of interactive storytelling, AI-generated worlds, and the belief that the best stories are the ones you help create.

What it does

DreamScape is a choose-your-own-adventure platform with three core experiences:

Create — Creators pick a genre, write narration for each scene, set branching choices, and generate 3D worlds in real time using World Labs. The platform handles the world generation; the creator focuses on the story. When they're done, they preview their story in XR and publish it to the community gallery.

Play — Players browse community stories in a gallery, pick one, and step inside. They walk through AI-generated 3D worlds while a narrator guides the story through a floating spatial dialogue panel. At key moments, the player makes choices that determine the character's fate — and each choice loads an entirely new world.

Remix — Finished a story? Fork it. Change the choices, add new branches, generate new worlds, and publish your version. Every remix credits the original creator, building a living ecosystem of interactive stories.

Our demo story, "The Interview," follows a woman who arrives at a corporate office for a job interview — only to discover the real interview happens in surreal liminal spaces. Three choices. Two endings. The interview is rigged against you. But the bold survive.

How we built it

DreamScape is built as a WebXR progressive web app designed for PICO headsets via WebSpatial.

World generation: We used the World Labs API (Marble 0.1-mini) to generate 3D Gaussian splat environments from text prompts. Each scene in the story is a unique AI-generated world — 8 total for "The Interview" demo.

Rendering: Spark.js 2.0 renders Gaussian splat (.spz) files directly in the browser with Three.js, enabling smooth 3D navigation through the generated worlds.

Spatial UI: WebSpatial powers the floating UI panels — the dialogue panel, choice buttons, and story gallery all exist as spatial elements in 3D space alongside the worlds.

Story engine: A custom React-based state machine tracks the player's current scene, choices, and a bold/safe scoring system that determines the ending. The entire story structure — narration, choices, reactions, loading text — is defined as data, making it easy for creators to build new stories.

3D character: A Mixamo character serves as the receptionist — physically present in the lobby scene as a 3D animated model, and as the narrator voice throughout the liminal spaces.

AI agent: OpenClaw handles story narration generation and text-to-speech voice for the narrator character.

Challenges we ran into

[FILL IN AFTER THE HACKATHON — here are likely challenges to reference:]

• World Labs generation takes 30-45 seconds per world, which we had to design around with cinematic loading screens that feel intentional rather than like waiting
• Rendering 8 unique Gaussian splat worlds while keeping performance smooth on a headset required careful attention to splat counts and LOD
• Building a creator tool AND a playable experience as a solo developer meant making hard prioritization calls throughout the hackathon
• WebSpatial panel positioning in 3D space required iteration to find the right balance between visibility and immersion
• Integrating a 3D character model with animations alongside Gaussian splat environments presented rendering pipeline challenges
• Scoping a platform vision into a demoable hackathon MVP while still communicating the bigger picture

Accomplishments that we're proud of

[FILL IN AFTER THE HACKATHON — here are likely accomplishments to reference:]

• Built a working choose-your-own-adventure that generates entirely new 3D worlds based on player decisions
• Created a creator tool that lets anyone build interactive stories with AI-generated worlds — no 3D modeling skills needed
• Designed a story ("The Interview") with 7 unique paths, 2 endings, and a scoring system that makes the interview feel rigged — because it is
• Turned World Labs' 30-45 second generation time into a feature, not a bug, with cinematic transitions featuring the narrator's dark humor commentary
• Demonstrated the full Create → Play → Remix loop as a solo developer in under 24 hours
• The receptionist character works both as a 3D presence in the lobby and as the omnipresent narrator voice — tying the whole experience together

What we learned

• World models are genuinely powerful for interactive storytelling — the ability to generate a unique 3D environment from a text prompt in under a minute changes what's possible for narrative experiences
• Designing around AI generation time is a creative challenge, not just a technical one — the loading screen became one of the most memorable parts of the experience
• WebSpatial makes spatial UI surprisingly accessible for web developers — floating panels in XR feel natural once the positioning is right
• The biggest design decision wasn't technical — it was choosing what to build for the hackathon vs what to show as a vision. The platform is bigger than any one demo.
• Solo hackathons require ruthless prioritization — having a clear "must ship / should ship / nice to have" framework saved the project multiple times
• AI coding assistance (Claude, Cursor, Windsurf) is a game-changer for designer-developers — the gap between "I can design this" and "I can build this" is shrinking fast

What's next for DreamScape: Interactive Story Worlds

Creator tool expansion — Full visual story tree editor, template library, genre-specific world prompt suggestions, and a "Story DJ" mode where creators set a vibe and the AI adapts narration, worlds, and music to match.

Community features — Story ratings, comments, creator profiles, trending stories, and a remix graph showing how stories fork and evolve across the community.

Multiplayer — Co-op storytelling where two players walk through the same story but make different choices, seeing how their paths diverge and reconnect.

Voice interaction — Replace button-based choices with natural voice input — just tell the narrator what you want to do, and the story adapts.

More story formats — Beyond choose-your-own-adventure: mystery investigations, escape rooms, guided tours, educational experiences — all built on the same AI world generation platform.

Mobile + headset — Optimize for standalone headsets (PICO, Quest) while maintaining the web-based approach for maximum accessibility. Any device with a browser can play DreamScape stories.

Built With

• three.js
• webspatial
• worldlabs