CozyClay
1.7k Try it
0000 / 0720 · 24fps

Block the shot in 3D, then generate it

Free, open source previs in your browser. Stage the scene, pose characters, set the camera and cuts, then hand the same shot to an AI video model.

Try it in your browser
Why

A text prompt cannot hold a whole scene

AI video costs money per take, and one miss means paying again.

Make those decisions first, in 3D, where they are cheap to change. The video model then gets the shot as a first frame, a greybox reference clip and a prompt that describes the real framing. It follows the blocking instead of guessing.

  • Where everyone stands Stage
  • How they move Move
  • Where the camera is Shoot
  • When it cuts Shoot
How it works

Stage, move, shoot, export

One local workspace that handles like a 3D editor, from the empty deck to the files a video model reads.

Step 1

Stage

Build the set from primitives. Add rigged characters, or capsule stand-ins.

A whole street from cubes and cylinders, railings and lamps included.
Step 2

Move

Describe the action. The compiler bakes every frame; QA checks collisions, foot slide and reach.

Ask the agent; the character walks to the ball and pushes it with both hands.
Step 3

Shoot

Shots on a timeline, dolly rails and cranes, a look through the lens.

The camera rides a rail while the crane takes it up, previewed live.
Step 4

Export

Everything the video model needs, in one pack. Every edit is one undo step.

  • Keyframe packFirst and last frame, greybox clip, camera, prompt
  • VideoThe shot as an mp4
  • PassesDepth and normal
  • Contact sheetThe shots side by side
  • OTIO cut listThe edit, for your editing software
  • StoryboardPanels with captions and an animatic
Direct it with AI

Your agent builds the scene while you watch

The same scene can be driven from three places, live in the viewport. Bring ChatGPT, Anthropic, OpenAI, Gemini, MiniMax, OpenRouter or CLIProxyAPI.

Agent panel

Chat beside the viewport and watch each change land. Sign in with ChatGPT or use your own key. Every change the agent makes is one undo step.

Open the studio, then press Cmd/Ctrl + B

MCP server

Add CozyClay to Claude Desktop, Claude Code, Cursor or any MCP client. The agent edits the scene in the studio you have open.

{
  "mcpServers": {
    "cozyclay": {
      "command": "npx",
      "args": ["-y", "cozyclay", "mcp"]
    }
  }
}

Terminal

Coding agents drive a running studio from the shell, JSON in and out.

npx cozyclay
cclay live status --wait
Quick start

One command

Node.js 22.19+ and a Chromium browser. The studio opens at 127.0.0.1:5180/app/.

npx --yes cozyclay@latest --via site

Never used a terminal? .

Asking an AI agent to set it up? Give it the repository link; its instructions are in INSTALL_FOR_AGENTS.md.

Full control list
InputAction
Right-dragLook around (fly)
RMB + WASDWalk while flying
RMB + Q/ECrane down / up
Middle-dragPan
Alt + dragOrbit the selection
ScrollDolly
W / E / RMove / rotate / scale tool
Ctrl/Cmd+Z, Ctrl/Cmd+Shift+ZUndo / redo
Ctrl/Cmd+DDuplicate the selection
FFrame the selection
Guides

Read

Greybox to AI video

Why a grey blockout is the most reliable input for a video model, and the block-then-generate loop.

Seedance 2.5 camera control

Getting the same camera on every take with a reference clip.

Previs software compared

Free, open source and browser-based previs tools side by side.

Camera control for AI video

Why camera prompts drift and what the models actually accept.

Block your first shot

Free, open source, and running in your browser in under a minute.

Try it in your browser Join the Discord

Install CozyClay

Run the Studio on your computer with the AI agent. Set up Node.js once, then launch CozyClay in your browser.

Step 1 of 2 · Install Node.js

Install Node.js

Download Node.js, open the .pkg file, and finish the installer.

Download Node.js

Already have Node.js 22.19 or newer? Continue to launch.