Topview
    • MCP/Skill
    • Plugin
    • API
    • Pricing
    1. Home
    2. World Labs Atlas

    AI Ads

    AI Video AgentAI Ads VideoAI Product VideoAI UGC VideoURL to Video

    AI Avatar

    AI Avatar GeneratorProduct AvatarDesign My AvatarAI Lip Sync

    AI Video

    AI Video GeneratorDrama StudioAI Video Body SwapAI Video UpscalerAI Video Watermark RemoverTikTok Watermark RemoverSora Watermark RemoverSubtitle RemoverAI Motion ControlAI Lip SyncURL to Video

    AI Image

    AI Image GeneratorAI Face SwapAI Image Character SwapAI Image UpscalerAI InpaintAI Image Text EditorPhoto Angle EditorAI RelightingAI Product PhotographyAI Virtual Try-OnAI Storyboard GeneratorExtract Color PaletteImage to Prompt

    AI Voice

    VoiceoverAI Voice CloningAI Music Generator

    3D

    3D World Generator3D Shot Composer

    Use Cases

    AdvertisingAffiliate MarketingEcommerceDTC BrandsAI Live Stream

    Resources

    BlogAffiliate ProgramLearning CenterAlternativeAPI

    Company

    Topview StudioPrivacy PolicyTerms
    Topview

    © 2026 TOPVIEW PTE. LTD.

    Singapore: 20 Collyer Quay #20-03, Singapore 049319
    Los Angeles: 15970 Los Serranos CC Dr #251, Chino Hills, CA 91709

    World Labs Atlas:The World's First Multimodal World Model

    Atlas is World Labs' next-generation omni world model. It brings camera-controlled generation, spatial reconstruction, space-time simulation, and explicit 3D output into one architecture built around a shared spatial context.

    Reference images

    Upload

    Describe your video idea

    Coming Soon

    What is Atlas?

    Generate, reconstruct, and simulate worlds

    Atlas reasons about visual context, camera position, and depth together.

    Official Atlas showreel by World Labs

    Atlas grounds images and depth maps to camera positions, then uses that spatial context to create new views, rebuild geometry, and simulate movement through a scene.

    01

    Generate

    Create new frames and videos along specified camera positions and trajectories.

    02

    Reconstruct

    Turn sparse images or video frames into novel views and explicit 3D representations.

    03

    Simulate

    Model space and time for reframed VFX shots and Real-to-Sim robotics workflows.

    Core capabilities

    What World Labs demonstrated with Atlas

    The launch material spans creative production and physical simulation. These are World Labs demonstrations and claims, not independent Topview benchmarks.

    Camera-controlled generation

    Atlas accepts camera geometry as a native input instead of relying only on words such as pan, crane, or orbit. World Labs showed hand-designed camera paths through a coherent generated scene.

    Demo: up to 1 minute at 1440p

    Spatial reconstruction

    Start with one or more images, generate unseen viewpoints, estimate geometry, and add more views when a faithful real-world reconstruction matters more than plausible invention.

    From sparse views to explicit 3D

    Space-time simulation

    Reconstruct motion from a small set of ordinary cameras, reframe an event from new angles, or generate the RGB and depth observations a simulated robot would see.

    VFX, navigation, and manipulation

    Image and 360 generation

    World modeling is the main focus, but World Labs also demonstrates prompt-following image generation, text rendering, varied visual styles, and 360-degree images from text or image prompts.

    Every frame is a view into a modeled world

    How the model works

    Spatial context turns references into positioned evidence

    Atlas processes a multimodal sequence one element at a time, with image and depth inputs tied to explicit camera poses.

    The practical idea is simple: each additional real view tells the model more about the scene. When less is observed, Atlas has more room to imagine; when more is observed, the reconstruction can rely on more evidence.

    1. 01

      Collect the context

      Provide text, one or more images, image sequences, camera poses, or depth information depending on the task.

    2. 02

      Ground views in 3D

      Place each frame at a known camera position so the model can relate appearance to space.

    3. 03

      Generate what comes next

      Autoregressive prediction and diffusion produce new multimodal elements conditioned on everything already in context.

    4. 04

      Render or reconstruct

      Return novel views, controlled video, depth, point clouds, or a Gaussian Splat scene for downstream work.

    Where Atlas could matter

    Creative and physical-world workflows share the same spatial core

    Atlas is positioned as a general foundation for scenes that need more than a single attractive frame. Its value is the connection between appearance, geometry, viewpoint, and change over time.

    VFX and virtual cinematography

    Design exact camera paths, create new views around a performance, and use reconstructed depth or 3D structure to support reframing and compositing.

    Previsualization and spatial design

    Arrange references as anchors in space, explore transitions between ideas, and test how a camera moves through a proposed environment.

    Games and interactive worlds

    Generate explorable scene representations and multiple consistent viewpoints that can inform environment concepts, layout, and prototyping.

    Robotics Real-to-Sim

    Reconstruct a real environment, vary its conditions, and generate simulated sensor views for navigation or manipulation training and evaluation.

    Images, panoramas, and long shots

    Use the same world-aware context for stylized images, 360-degree scenes, and longer camera-controlled sequences.

    Atlas vs Marble

    A foundation model announcement is not the same as a public creation product

    Atlas and Marble are related, but they are not interchangeable names. This distinction matters when deciding what you can use today.

    Dimension
    World Labs Atlas
    Marble on Topview

    Primary role

    World Labs Atlas

    A next-generation omni world model for generation, reconstruction, and simulation across creative and robotics tasks.

    Marble on Topview

    A currently available model family for generating and exploring 3D worlds in Topview Canvas.

    Availability

    World Labs Atlas

    Early access with select World Labs partners; public pricing and general access have not been announced.

    Marble on Topview

    Marble 1.1 and Marble 1.1 Plus are selectable in Topview's 3D World workflow.

    Inputs

    World Labs Atlas

    World Labs describes a shared context spanning text, images, camera poses, depth, and video represented as image sequences.

    Marble on Topview

    Topview accepts a text prompt plus one image or one short video reference; image and video references are mutually exclusive.

    Outputs

    World Labs Atlas

    Demonstrated outputs include frames, video, depth, point clouds, and 3D Gaussian splats.

    Marble on Topview

    Topview returns an explorable Gaussian Splat world card and lets creators capture views back into Canvas.

    Relationship

    World Labs Atlas

    World Labs says Atlas will power future versions of Marble and other World Labs products.

    Marble on Topview

    The current Topview integration remains bound to Marble 1.1 and Marble 1.1 Plus, not Atlas.

    Topview is an independent product and is not the official World Labs website. Availability can change, so verify the live model selector and World Labs' official channels before planning a production dependency.

    Technical details

    Atlas model architecture and published scope

    A compact view of what World Labs has documented publicly, without filling gaps with assumed model sizes, pricing, or API parameters.

    Details below reflect World Labs' September 1, 2026 announcement and may change during early access.
    Architecture
    Multimodal autoregressive diffusion transformer; diffusion is implemented as a rectified flow model.
    Current input types
    Text, images, camera poses, and 3D depth maps; video is represented as a sequence of images.
    Spatial control
    Images and depth maps are conditioned on explicit camera poses inside a shared spatial context.
    Published video demonstration
    World Labs shows up to one minute of 1440p video generated from a small set of references along a hand-designed camera path.
    Reconstruction range
    One to dozens of input images are highlighted; World Labs also states the context can use more than one hundred views.
    Explicit 3D outputs
    Per-frame depth, 3D point clouds, and Gaussian splats, alongside generated image and video frames.
    Release status
    Select-partner early access. No public Atlas pricing, general API documentation, or downloadable weights were published with the announcement.

    Frequently asked questions

    World Labs Atlas FAQ

    Straight answers about the model, its inputs and outputs, Atlas access, and what Topview supports today.

    Third-party model and brand names are trademarks of their respective owners. Topview provides access to supported AI models through its independent platform and is not the developer of those third-party models.

    Create with what is available today

    Explore a 3D world now with Marble in Topview Canvas

    Atlas remains in select-partner early access. For a live creator workflow today, use Marble 1.1 or Marble 1.1 Plus to generate an explorable world and capture viewpoints back into Canvas.

    Open the Marble 3D World PageReview Atlas on World Labs

    The Topview destination uses current Marble models and does not imply Atlas access or affiliation with World Labs.