A 3D scene for directing image and video generation in Intangible 2.0

The studio combines a 3D scene, editable cameras, and external generative models. Its browser canvas brings agents, references, and shots together.

The set, objects, and cameras are retained in a 3D scene that can be modified between generations. This principle shapes the workflow in Intangible, whose cofounder Bharat Vasan presents version 2.0 as a studio combining images, video, and 3D. A canvas brings together references, scene elements, shots, and renders, with agents handling some of the operations.

The offering focuses on project planning and visualization: film sequences, advertising campaigns, products, retail spaces, or events. The interface runs in the browser. The scene provides a shared foundation for multiple shots, while final images and videos are produced using the generative models selected by the user. From the brief to the cameras, with editable elements

The studio documentation describes a workflow that begins with an instruction and files: photos, PDFs, or 3D objects. Agents ask questions, write a brief for approval, prepare references, and assemble a scene. They then place a camera for each shot before generating images and videos.

The work remains accessible through three complementary modes. Build is used to compose sets and objects; Compose handles shots and animation; Visualize produces renders. A camera’s position, framing, and movement can be defined in the scene rather than described solely in text. Users can move between conversation and manual adjustments.

The canvas also preserves relationships between elements: a render can be traced back to the shot and scene it came from. This organization makes it easier to revisit a project, since a change can target an identified object, camera, or reference. It provides a shared framework for iteration without establishing that every generative output will perfectly preserve all its details. What comes from the studio and what comes from third-party models

The model catalog documents the use of external providers. For images, it includes models from Black Forest Labs, Google, OpenAI, and ByteDance. Video options include families such as Veo, Kling, Luma Ray, Seedance, Wan, and Runway. Choices vary by generation type, supported references, duration, and output resolution.

Model aggregation is therefore part of the product. The studio-specific layer covers the editable scene, relationships between objects, cameras, and preparation of instructions for the generators. The term “wrapper,” when used to describe an interface providing access to third-party models, captures that component without encompassing the full set of documented composition tools.

The company describes its Spatial Intelligence Engine as a semantic layer connecting objects, spaces, and relationships through a proprietary knowledge graph. The pages reviewed describe its uses without providing a complete technical architecture that would allow the technology to be evaluated separately. The founder’s claim of “up to 10x fewer throwaway generations” is not accompanied by a methodology or sample size in the sources reviewed. Importing geometry does not preserve the entire original project

The 3D model import rules specify several limitations relevant to production. Supported formats include GLB, glTF, FBX, OBJ, and DAE. Geometry is preserved, but textures are removed, except for transparency masks, and materials become white. Visual references are used to restore the appearance at render time.

Imported animations are not preserved. However, some rigged models retain their structure and can be posed in the studio. For static objects, the hierarchy is flattened and meshes are merged: parts that need to move independently must be imported as separate files. These constraints matter for teams looking to reuse a product or character already prepared in another application. Connecting an agent and choosing usage rights

An MCP server provides canvas access from compatible clients such as Claude, ChatGPT, or Cursor. Presented uses include building scenes, placing cameras, and creating renders. This connection expands the entry points into a project; composition tools remain accessible in the studio.

The pricing page distinguishes the rights associated with each plan. The free tier lists a one-time allocation of 150 credits, limited models and projects, and image downloads. MCP access, 3D import, scene export, video downloads, and commercial use are included in paid plans.

Explorer is listed at $25 per month, billed annually, with 500 monthly credits. Business is listed at $50 per user per month, billed annually, with 1,000 monthly credits per user that can be pooled. The number of images or seconds of video produced depends on the model and resolution: the displayed equivalents remain approximate.