Real-Time Generative Video

World Labs Marble

World Labs Marble creates, edits, expands, and exports persistent 3D worlds from text, images, video, and layouts.

Visit official website
World Labs Marble official website first-screen screenshot, captured September 5, 2026.
Official website screenshot · · View source

Editorial verdict

A world-generation workflow for persistent spatial assets, not a continuously generated interactive video feed.

Best for: Artists, designers, and developers who need a generated 3D world to edit, export, or explore after creation

Editorial notes

World Labs Marble at a glance

World Labs Marble is a useful boundary case for a real-time-generation directory. It makes persistent, high-fidelity three-dimensional worlds from text, an image, multiple images, video, or coarse 3D structure. Users can then edit, expand, combine, export, and explore those worlds. That makes it relevant to live-video production, simulation, XR, and interactive environments when a generated spatial asset feeds a later experience. It is not, however, a generator that continuously redraws a video feed while a viewer interacts. Marble belongs here as a workflow integration, not as native or near-real-time live video.

The official documentation is candid enough to make the distinction. Marble estimates roughly 30 seconds to create a panorama from text, an image, or a 3D structure; roughly two minutes for a panorama from multi-image or video input; around 20 seconds for a draft; around five minutes to create or expand a world; and around an hour for a high-quality mesh. These are backend-processing estimates for named operations. They are not model inference benchmarks, delivered-video frame rates, queue commitments, or end-to-end response guarantees. They make a good planning baseline precisely because they are slow enough that a production should use pre-generation rather than pretend it can react live.

Choosing a model and access path

Marble offers several models with a clear speed-and-cost tradeoff. The documentation calls Marble 1.0 Draft its fastest model, intended for quick idea exploration and prompt testing, while Marble 1.1 is recommended as a general starting point and Marble 1.1 Plus is meant for larger worlds. The label “fastest” is qualitative product language; the approximately 20-second draft estimate supplies the more meaningful time context. A team can use drafts to narrow an art direction, then run a higher-quality world generation only after an approved input is ready.

The Marble web app has a Free plan with up to four world generations and supports text, image, multi-image, and 360 panorama inputs. Standard adds video input, exports, and Draft mode; Pro adds expanded worlds, enhanced rendered-video outputs, higher-quality mesh export, and commercial rights. These are official plan limits at verification time, not assumptions about API usage. The API is a separate product path: its documentation requires a World Labs Platform account, a payment method, purchased API credits, and an API key. Marble subscription credits cannot be used as API credits. That separation is important for both a prototype budget and an implementation handoff.

The World API supports Marble 1.1, 1.1 Plus, 1.0, and 1.0 Draft model values and accepts a task that can produce a world from text, images, multiple images, or video. It returns a world record and assets after the operation completes. That is a normal asynchronous generation workflow. Build a status UI, preserve an input-to-world audit trail, and let a user revisit a completed world instead of blocking a live interface waiting for a spatial result.

Operational fit

Marble can produce useful material for a live production without becoming the live production engine. A broadcaster might generate a virtual set beforehand, export a splat or mesh, and bring it into a renderer. A training team might make a space for a later interactive module. A creative department could produce a camera path or cinematic recording after art direction is accepted. In all cases, test the exported format, rights, target device, navigation performance, and render pipeline before committing it to a show.

Mobile access is available, but the documentation recommends desktop for the full feature set because some advanced creation, editing, 3D-structure, and panorama capabilities are not supported on mobile. Add that constraint to a production brief. Also consider input rights, any sensitive imagery in source video, world sharing permissions, and the retention of generated assets.

Choose Marble when the desired result is a persistent spatial world that will be refined or explored after it is generated. Choose PixVerse R1 for an actively changing world during a session, Decart for live transformation of an incoming feed, and MiniMax H3 Max or Runway for short rendered video assets.

Capabilities

  • Text-to-world
  • Image-to-world
  • Video input
  • 3D layout input
  • World editing
  • API access

What real-time means

Workflow: Marble publishes backend generation estimates from about 20 seconds for a draft to about five minutes for a world, not continuous frame generation.

Access and pricing

Pricing model
Free and paid credit subscriptions for Marble, with separately billed World API credits.
Free access
free
API availability
yes
Real-time capability
workflow

Visit website for current plans

Pros

  • Persistent 3D output
  • Multiple input forms
  • Published generation estimates

Cons

  • World creation takes minutes
  • API credits are separate
  • Full editor features favor desktop

Official sources

  1. Source 1
  2. Source 2
  3. Source 3
  4. Source 4

Last verified:

Alternatives

Real-Time Generative Video

PixVerse R1

A native real-time candidate for interactive worlds, shared sessions, and live visual control—not ordinary clip generation.

Best for: Teams prototyping interactive worlds, live media, simulation, XR, or shared visual experiences

View details

Real-Time Generative Video

Decart Lucy

A native live-transformation system for teams that need generative changes to an existing video stream as it runs.

Best for: Enterprises, brands, creators, and developers prototyping live video transformation and interactive media

View details

Real-Time Generative Video

MiniMax H3 Max

A fast short-clip option when faster-than-playback iteration matters more than a continuously interactive video stream.

Best for: Creative teams and developers testing short prompt, image, and reference-conditioned clips quickly

View details

Real-Time Generative Video

Runway

A practical short-clip generation workflow whose fast models improve iteration but do not create a continuous live video stream.

Best for: Creative teams making image-conditioned short video takes with browser, mobile, or API workflows

View details