OpenArt Worlds Lets Creators Walk Inside AI-Generated 3D Scenes

The way AI creative tools have worked until now is simple: you prompt, the model generates, you take it or leave it. OpenArt Worlds is trying to break that loop entirely. Powered by the World Labs API, it lets creators upload a single image and walk into a persistent, navigable 3D environment they can return to, reframe, and build within over time -- more like a virtual film set than a one-shot generation.

From prompt roulette to spatial direction

The core problem OpenArt Worlds is solving is one every AI creator knows: you describe what you want, the model gives you something close, and then you prompt again, and again, hoping to land on the right shot. There is no way to hold the scene still and just move the camera. OpenArt Worlds changes that.

For the first time in AI content creation, creators are no longer at the mercy of what the model decides to generate. OpenArt Worlds lets you build a fully navigable 3D world from a text prompt or image, walk through it freely, and take the shot from exactly the angle and composition you want. Wide establishing shots, top-down views, over-the-shoulder framing -- all from the same environment, without regenerating anything.

The workflow breaks down into three steps:

1. Build: Upload any image -- a photo, an AI-generated piece, or a concept painting -- and generate a navigable 3D world.
2. Direct: Move freely through the environment, set camera angles, and frame the exact shot you want.
3. Capture and Compose: At the moment of capture, creators can add characters, objects, or any creative element via prompt, outputting a production-ready 2D image.

OpenArt Worlds is a persistent AI creation space that converts images or text prompts into explorable 3D scenes. Unlike standard AI image generators that produce one-off outputs, Worlds saves every environment permanently inside your library. Come back next week, add a new character, capture a different angle -- the world is still there.

The technology underneath

The spatial engine behind this is Marble, World Labs' large world model (LWM -- a class of AI model designed to reason about and generate three-dimensional space, as opposed to flat images or text). Marble uses 3D Gaussian Splatting, a technique that balances photorealism and computational efficiency. Traditional 3D generation is either low quality or so heavy it is unusable in real-time. Gaussian Splatting represents a scene as millions of small, overlapping ellipsoids rather than polygons, which makes rendering fast and editable.

The World API makes it possible to create navigable spatial environments directly from text, images, panoramas, multi-view inputs, and video. These worlds can be rendered on the web, exported into downstream tools, or integrated into interactive systems and simulations. World generation runs asynchronously, so there is no proprietary tooling required on the integration side.

The division of labor in the OpenArt integration is clean:

• World Labs handled: world generation, spatial reconstruction, and persistent environment consistency
• OpenArt built: camera navigation, shot capture workflows, prompt-based composition, and export pipelines

Coming soon: export to Gaussian splat, 3D mesh, Unreal Engine, and Blender formats. Access is available inside OpenArt Suite using OpenArt's credit system, with a free tier of 50 credits per month and paid plans from $7 per month.

Who is World Labs, and why does this matter now

World Labs is Fei-Fei Li's first commercial venture, co-founded in 2023 with researcher Justin Johnson, on the thesis that the next major AI capability gap is reasoning in three dimensions. Li is widely known as the "godmother of AI" and the architect of ImageNet, the dataset that kicked off the modern deep learning era.

World Labs secured a $200 million investment from Autodesk as part of a larger $1 billion round from backers including AMD, Emerson Collective, Fidelity, Nvidia, and others. The company, which emerged from stealth in 2024 with $230 million at a $1 billion valuation, declined to say whether the latest round boosted its valuation. However, reports suggested it was aiming to raise at a $5 billion valuation.

The partnership between World Labs and Autodesk will see the two companies collaborating to explore how World Labs' models can work alongside Autodesk's tools, starting with a focus on entertainment use cases. That is a meaningful signal: Autodesk's software underpins architecture, engineering, and film production pipelines globally.

The World API launched in January 2026, and OpenArt was among the first consumer-facing platforms to build on it. Early partners also include Preview, a workspace for filmmakers, and Fenestra, an architectural visualization tool.

The numbers that tell the real story

OpenArt Worlds became one of OpenArt's biggest launches of the quarter. The case study numbers are striking:

• 7.3 million impressions on X at launch
• 4,100 likes and 1,600 reposts
• 770 X News posts within the first 9 hours
• 250+ top creators posting about OpenArt Worlds across X, Instagram, and YouTube
• 12 million impressions from influencer campaigns alone
• Trending placement in X News in the US

But the more durable signal is behavioral: creators began building thousands of worlds every week and returning to saved environments to capture new shots -- the kind of retention that suggests the tool is actually changing how people work, not just generating viral clips.

Who wins, and what opens up

The clearest winners here are independent creators and small studios who previously had no path into persistent 3D production. For decades, building a 3D world required a team: a modeler to create geometry, a texture artist to paint surfaces, a lighting designer to set the mood, and a level designer to make it feel real. Even a simple room could take days. OpenArt Worlds compresses that to minutes, with no 3D skills required.

For World Labs, the OpenArt integration is a proof point for the API business model: let other platforms handle the creator-facing UX while Marble handles the spatial intelligence underneath. OpenAI and Google DeepMind have added image and video generation to language-centric platforms; World Labs is betting that persistent, navigable 3D environments require a separate model architecture and constitute a product category those platforms cannot fully address.

The competitive landscape is real, though. World Labs' two most comparable direct competitors are Luma AI and Decart, both venture-backed startups building generative models for 3D or interactive environments. Most competitors, like Google's Genie, focus on real-time, on-the-fly generation -- creating interactive environments that are generated frame-by-frame as you explore. While impressive, this often leads to inconsistency, visual artifacts, and the inability to export a stable, final asset. Marble's persistent world approach is a deliberate architectural choice against that pattern.

The second-order implications reach beyond creative tools. World Labs frames spatial intelligence as essential for robotics and simulation: training robots requires vast amounts of data from diverse environments, and world models can generate unlimited synthetic training worlds. The World API already integrates with NVIDIA Isaac Sim, MuJoCo, and RoboSuite. The same model powering a creator's virtual film set is also being used to train robots.

What comes next

OpenArt has a clear roadmap from here. The next major step is direct character placement inside the 3D world itself, allowing creators to position characters spatially before capturing the shot. That would close the loop on the full virtual production workflow: build the world, place the characters, direct the shot, export the frame.

Li has said spatial intelligence models could be put to use in the future for augmented and virtual reality or robotics. The OpenArt case study is the clearest public demonstration yet that the World Labs API is production-ready and scaling -- and that the shift from generating images to building worlds is already underway.