Autodesk brings 3D Editor + Canvas to Flow Studio: generative video scenes are assembled and framed in 3D

Autodesk has released 3D Editor + Canvas for Flow Studio — a workspace where a scene is generated, assembled, rendered and refined in one place. The company announced the launch on August 4, 2026; the pairing is included in all pricing plans, including the free one.

Flow Studio is the platform previously known as Wonder Studio: it was created by Wonder Dynamics, which became part of Autodesk. The company calls the update itself a major expansion of the platform and a step toward an AI-native creative environment that brings together generation, directing, scene assembly, rendering and refinement. Wonder Dynamics co-founder Nikola Todorovic explains the idea this way: “We live in 3D and we shoot in 3D, but most generative video tools still force creators to perform and edit in a 2D world. Trying to direct a performance or design a camera move through a prompt or a drawing on a flat image is an inherently limited approach.”

On the befores & afters podcast, Todorovic puts the emphasis precisely on control: the point of the update is that the creator gains more power over the scene — they can build the staging, set camera movement and timing, and then refine the result.

3D Editor: staging a scene in a three-dimensional environment

In 3D Editor a scene is built in a realistic 3D environment: characters are placed, the staging is defined, cameras are animated with keyframes, and timing, pacing and camera movement are controlled. A single scene brings together characters, AI-generated worlds and environments, animations, camera tracks, live-action elements and AI motion-capture data — that is, shot footage and generated content end up in a shared space.

Environments can be pulled in from third-party tools — Autodesk names Marble from World Labs. According to the company, a finished cinematic scene can be built on that basis in minutes. No 3D experience is required: the scene is assembled from ready-made generated assets, and when speed matters more, you can stay in the 2D mode of Canvas.

Canvas: a node-based 2D workflow for rendering and refinement

Canvas is the second half of the pairing, a node-based 2D environment. The shot assembled in 3D Editor is rendered and refined here: variants are generated, explored and iteratively refined in the node graph, with rendering and refinement running in parallel. The level of control is chosen for each shot: 2D for speed and simplicity, 3D for precise work with composition, camera movement and scene assembly.

Generation is handled by third-party models — Autodesk names Veo, Seedance, Kling, Gemini, Omni, Nano Banana and Marble (World Labs); according to the company, the set of available models is expanding.

Who can access it

3D Editor + Canvas are included in all five Flow Studio plans — Free, Lite, Standard, Pro and Enterprise. Autodesk publishes current prices and a plan comparison on the purchase page. The update is aimed at independent directors, studios, artists, marketers, agencies and storytellers, as well as experienced 3D specialists who need direct control over scene assembly and framing.

As for plans, Autodesk names possible future capabilities: expanded integration with Autodesk Maya and other DCC packages, additional asset import options, scene assembly with the help of AI, and agentic directing scenarios.

Why Autodesk is moving generative video into 3D

The launch of 3D Editor + Canvas looks like a response to production's main complaint about generative video — the lack of controllability. A shot obtained from a text description is hard to reshoot from a different angle, to match with the neighboring shot, and to carry through several rounds of notes from the director and the client. Autodesk itself frames this as a shift in the task: generating content has become easy, the difficulty now is controlling it — the same subject, according to the company, came up in the conversation about authorship and creative intent at the Cannes Film Festival.

A scene with cameras, keyframes and persistent assets brings the familiar studio discipline back into this process — the same discipline Maya, 3ds Max and Arnold are built on, and Autodesk speaks about that continuity directly. Judging by how the product is put together, the bet is on a control layer on top of third-party generative models: models are plugged in and swapped out, while the scene, the asset library and the connection to familiar pipelines remain the value. This likely embeds Flow Studio into the company's broader lineup of cloud tools.

The flip side of such an architecture is dependence on external model providers: a change in the terms, prices or availability of any of them directly affects users. Todorovic himself also speaks on the podcast about caution regarding machine learning in VFX and CG — the question of the limits of neural networks in production remains open. Against this backdrop, competition in generative video will likely shift from the quality of an individual shot to the repeatability of the result and integration with existing pipelines, while the skills of framing a shot and working with a camera retain their significance — what changes is the environment in which they are applied.