Topview open-sourced 3D Builder: plan the scene before generating AI video

On October 15, 2026 Topview AI published a DEV.to write-up and opened the topviewai/topview-3d-builder repo: a local tool to build an editable 3D scene and reference frames first, then send the job to a video generator.

01

Why previs before generation

Spatial mistakes — who stands where, where the camera looks — often show up in a pure text prompt only after you pay for a generation. 3D Builder moves that failure to a cheap step: nudge a character, change a camera preset, re-render a PNG.

Topview’s post separates spatial intent (layout) from visual generation (final look and motion). The tool owns the first; your video model owns the second.

02

What shipped and under which license

The topviewai/topview-3d-builder repository is Apache-2.0. Bundled characters, poses and listed assets use CC BY 4.0 with attribution. A Topview account is not required for local previs.

The stack includes agent skill/plugin manifests for compatible agents, a Python CLI (topview-3d-cli) and a local Studio/renderer on Chromium. Scene state lives in a .topview-3d/ directory inside the project.

The post frames an early release: limited characters and poses, static cameras, no full walk-cycle library. It is previs, not a full 3D animation suite — as the authors state.

03

A typical workflow

Init a project with the CLI, open Studio or ask an agent via the skill to build a scene from a text brief, batch-edit nodes and poses, run evaluate/inspect views, save PNG references and feed those into Kling, Runway or another generator.

Install notes in the post assume Python 3.11+, Node 20.6+ and a one-time Playwright/Chromium setup. Confirm the package version against the repo README — the DEV.to text cited topview-3d-cli 0.1.6 as an example.

  • project init → status → studio open
  • asset list / camera presets to learn the catalog
  • batch node and pose edits instead of «invent everything again»
  • evaluate and inspect views catch geometry and framing before expensive generation
04

Where it saves money — and where it does not

Previs cuts pointless rerolls from «character A occluded B» or «camera too high.» It does not guarantee a video model will perfectly honor a reference frame.

Topview does not publish a measured credit-savings percentage — and we will not invent one. Benefit depends on your model, per-second price and how often layout errors drive rerolls today.

Sending a frame into Topview Canvas is a separate optional signed-in step; the local builder itself works without a Topview account.

05

Fitting it into business video production

Vertical-volume teams place previs before master-clip generation: especially dialogue scenes, multi-object product demos and any shot where eyeline matters.

The master then moves into versioning and seeding — or into a single vertical ad. Making videos with AI on our side already includes scene selection and rebuilds; 3D previs is a sensible layer if you own the pipeline.

Generator comparisons sit in the 2026 guide to AI video models. If you want one tightly directed clip without running a CLI in-house — handcrafted production starts from $190.

frequently asked

01When did Topview open-source 3D Builder?
The public DEV.to write-up is dated October 15, 2026; the topviewai/topview-3d-builder repo is Apache-2.0.
02Do I need a Topview account?
Not for local previs and reference renders. Sign-in is only for optional cloud steps such as Canvas.
03Is this finished AI video?
No. It is an editable 3D layout plus PNG references. A separate model produces the final video.
04Will it save N% of credits?
The authors do not publish a measured percentage. Savings appear only if you actually cut layout-driven rerolls.
05What are the licenses?
Code and schemas are Apache-2.0; listed built-in assets are CC BY 4.0 with attribution.