Topview open-sources 3D Builder, an agent-built 3D stage for planning AI video shots
Topview 3D Builder is a free, open-source tool where your AI agent builds an editable 3D scene from a description or a reference image, so you can set blocking, props and camera moves before generating any video. Everything up to the video step runs on your own computer with no account, and you can take the renders to Topview or any video model.
What it does
You describe a shot to ChatGPT or Codex, Claude Code, Cursor, Qoder or another agent with skills support, and it builds the scene from built-in characters, poses and props and places the cameras. The scene is checked automatically for floating or clipping characters, spacing and framing, and you can keep talking to the agent or drag things around in its Studio.
You then plan push-ins, pull-backs, rises, orbits and cuts on a timeline and render several cameras from the same scene, so positions stay consistent from shot to shot.
Worth knowing
It needs Python 3.11+, Node.js 20.6+ and uv or pipx, and it's licensed Apache 2.0. The 3D scene fixes blocking, scale and camera angles, while texture, lighting and performance come from the video model; the example videos in the README were rendered on Topview.ai. Topview recommends Qwen 3.8 Flash as the agent's model, saying it costs about 1/150 of GPT 6 Astra for a similar result.
Try it yourself
Build a small café. Two people sit across a table from each other. Put the camera outside the window and slowly push in over four seconds. Render the first, middle and last frame.