Starting Today, Fei-Fei Li’s AI “World-Creation Tool” Is Available to Everyone — Build a 3D World in One Sentence | Hands-On Test Included
Marble — The Viral 3D World Generator Is Now Open to All
Yes — it's that 3D world generation model that went viral in the AI community just two months ago. Fei-Fei Li’s World Labs has now opened Marble to the public and released a full suite of powerful features in one go.
---
🔑 Key Features
- Multi-modal generation: Accepts text, single image, multiple images/video, or 3D layout as input.
- AI-native editing: Swap elements or adjust structure in fully generated 3D worlds.
- World expansion & merging: Seamlessly extend scene boundaries or stitch worlds together.
- Chisel mode: Build block-based frameworks, AI auto-fills style and details.
- Full-format export: Gaussian point clouds, triangular mesh, video — ready for game engines.

Low barrier to entry: Enter a prompt or upload a photo; AI generates a complete immersive 3D space. This could mean traditional 3D modelers may need to rethink their career trajectory…
🔗 Test Marble here: https://marble.worldlabs.ai
---
✍️ From Text to Fully Rendered Worlds
Let’s begin with a real example to see Marble’s text-generation power.
Example Prompt #1

Prompt: A lived-in hobbit kitchen, filled with woven baskets and copper kettles, bathed in tranquil pale blue sunlight.
🔹 Result: Warm-toned wooden furniture, copper utensils, rich lighting — all spatially coherent.
Example Prompt #2 — Complex Style Fusion

Prompt: A space station kitchen blending medieval tavern aesthetics with orbital technology, featuring checkerboard floors and stainless steel fixtures, glowing under soft aqua lighting.
🔹 Result: Accurate mash-up of medieval and futuristic styles.
Personal Test: Generated a forest scene with minor repeated moss textures, but tree bark details and mist layering were impressive.

---
🎨 Beyond Pure Text Input
Marble supports multiple creation modes:
- Single image → panoramic world
- Multiple images/video → stitched 3D space
- 3D layout → AI auto-fills detail

Single Image Test
- Uploaded a room photo
- Wait ~6 minutes
- Result shows edge distortion, but early-stage quality is strong.

Verdict: Multi-modal design puts Marble far ahead of most tools currently available.
---
✏️ Editing Features — A Game Changer
Instead of stopping at generation, Marble enables intelligent, spatially aware editing:
- Local replacements: e.g., remove a trash bin — AI updates scene lighting automatically.
- Structural changes: Modify walls, replace furniture layouts while respecting 3D space.


This is not flat image manipulation — it’s object-aware 3D editing.
---
🌍 Expanding & Combining Worlds
- World Expansion: Fill blurred scene edges with auto-generated content.
- Combination Mode: Merge multiple generated spaces.

💡 Example: Official demo stitched an entire train carriage sequence — shrinking concept-to-delivery timelines dramatically.

---
🛠 Chisel Mode — Style/Structure Separation
Chisel workflow:
- Build with basic geometric blocks (like LEGO).
- Describe desired style (modern art museum, Scandinavian bedroom).
- AI auto-renders detailed 3D world.
Style Variations with Same Structure
- Modern Art Museum: White walls, abstract art, curved sculptures.
- Nordic Bedroom: Warm wood furniture, glacier window views.


📌 This is akin to HTML (structure) + CSS (style) but for 3D modeling.
---
📤 Export Options
- Gaussian Splats: Highest fidelity, renders directly in browser.
- Triangle Meshes: Low-poly for collisions, high-poly for render; import into Unity/Unreal.
- Video: Pixel-level camera control + AI enhancement (smoke, fire, water effects).


---
🎯 Applications
- Games: Generate environments, import into engines, add interactivity.
- Film: Rapid scene previews for planning.
- VR: Load directly into Vision Pro or Quest 3 for immersive experiences.
---
💰 Pricing
Tiered Subscription Model:
- Free: 7,000 points/mo, 4 worlds, text/image input only.
- Standard ($20/mo): 20,000 points, 12 worlds, multi-image/video, Chisel, basic exports.
- Professional ($35/mo, 1st mo $1): 40,000 points, 25 worlds, expansion, video enhancement, high-res export, commercial license.
- Flagship ($95/mo): 120,000 points, 75 worlds, all features unlocked.

---
📈 Market Reception
- Garry Tan (YC CEO): Praised launch significance.
- Indie devs & VR creators: Applauded lowered barrier to 3D creation.
However, GDC Survey: One-third of devs see negative industry impact — citing IP concerns, energy costs, declining quality, layoffs.
---
🌌 Fei-Fei Li’s Vision — Spatial Intelligence
Core premise:
> LLMs excel at language, but lack 3D physical world understanding — distance, rotation, physical laws.
Human intelligence thrives on spatial reasoning — crucial for driving, design, scientific discovery.
Spatial Intelligence Roadmap:
- Near-term: Creativity (games, film, architecture) → Marble’s current domain
- Mid-term: Robotics training in simulated spaces
- Long-term: Science breakthroughs, drug development, education

---
🆚 Market Landscape
Other players: Decart (real-time gen), Odyssey (basic free trial), Google Genie (research preview).
Marble’s Differentiators:
- Persistent full 3D worlds (less distortion)
- End-to-end workflow (gen → edit → export)
- Commercialized model with free & paid tiers

---
🏁 Final Thoughts
From ImageNet (“What this is”) to Marble (“Where it is & how it interacts”), Fei-Fei Li’s mission remains constant: teach machines to see, understand, and act in the visual world.
Could this ignite the next AI paradigm shift? Undetermined — but technically compelling.
For creators: Marble + open-source publishing ecosystems like AiToEarn may form the backbone of future AI-driven content pipelines — enabling generation, multi-platform distribution, analytics, and monetization across Douyin, Kwai, WeChat, Bilibili, Rednote, Facebook, Instagram, Threads, YouTube, Pinterest, and X.

---
Would you like me to also add a comparison table between Marble’s subscription tiers and competing products so readers can quickly assess value? That could make the rewritten Markdown even more reader-friendly.