Cozy Clay: The Free 3D Previz Tool for Planning AI Video Shots
Cozy Clay is a free, open-source 3D previsualization tool for AI video. Here's what it does and how to run it through Pinocchio.
What is Cozy Clay?
Cozy Clay is a free, open-source 3D previsualization tool built specifically to help people plan shots for AI video generation. Instead of jumping straight into a text prompt and hoping a model like Kling, Veo, or Sora gives you the camera angle and blocking you want, you build a rough 3D scene first (clay-style characters, simple sets, a camera you can move) and use that as a reference for the final AI-generated shot. It runs in a web browser, which means no GPU requirements and no heavyweight local install beyond getting it running in the first place.
The core idea isn’t new. Creators have been using Blender or Cinema 4D to block out scenes, then feeding renders into video models like Seedance or Minimax as a workflow for getting controllable camera moves and consistent staging. That Blender-to-AI-video pipeline produces good results, but it’s slow, render times stack up, and if you’re not already fluent in Blender, getting the camera move or character pose you actually want can take a long time. Cozy Clay exists to cut that friction out by being purpose-built for previs rather than full 3D production.
Why does AI video need a previz step at all?
Seven tools to build an app. Or just Remy.
Editor, preview, AI agents, deploy — all in one tab. Nothing to install.
Text-to-video models are good at generating motion and texture, but they’re still inconsistent at giving you precise camera control, blocking, and spatial continuity across a shot. If you need a specific framing, a character crossing from screen left to screen right, or a camera push that matches a reference shot, typing that into a prompt and hoping for the best is unreliable.
Previsualization solves this the way it always has in traditional filmmaking: you build a rough version of the shot first, lock the camera and blocking, then use that as a guide. Applied to AI video, that means a 3D previz pass becomes the “storyboard with depth” that you feed into the generation model, either as a reference image, a sequence of frames, or a motion guide. The appeal of a tool like Cozy Clay is that it’s made for exactly this handoff rather than being a general-purpose 3D suite you have to bend to the task.
How do you install Cozy Clay?
The recommended path is running Cozy Clay through Pinocchio, a free local app that acts as a launcher for open-source AI models and tools. Pinocchio handles the download and setup of models across video, audio, and image generation, so instead of manually configuring dependencies, you install Pinocchio once and use it to pull in tools like Cozy Clay with far less friction.
A few practical notes if you’re setting this up:
- The initial download takes a while, so don’t expect it to be instant.
- Once installed, Cozy Clay runs locally in your browser, which keeps hardware requirements low since there’s no heavy local rendering happening the way there would be in Blender.
- Camera movement inside the tool uses right-click-drag plus WASD keys, which isn’t necessarily the first thing you’d try, so it’s worth knowing going in.
How does the actual previz workflow work?
Once it’s running, Cozy Clay gives you a 3D scene populated with clay-style characters and objects that you can reposition manually. Characters come with basic rigging, so you can pose or move them rather than just placing static models. Because everything in the scene is a real 3D object, you can rotate, reposition, and frame it from any angle, which is the entire point: you’re hunting for the camera position and composition you want before committing to a generation.
There’s also an automation layer. Cozy Clay supports hooking in an AI agent, such as OpenAI’s Codex, using your own API key or even a standard ChatGPT subscription connection. Anthropic keys work too if you want to drive the scene with Claude models instead. Practically, this means you can describe a scene in natural language (“build me a private investigator’s office” was one example demonstrated) and have the agent construct the set and layout for you, rather than placing every object by hand. From there you can still take manual control to fine-tune blocking, adjust the camera, or pose characters exactly the way you want before exporting that as your previs reference.
The combination matters: agent-driven scene building for speed, manual control for precision. You’re not forced to choose one or the other.
Is Cozy Clay worth using over a Blender workflow?
One coffee. One working app.
You bring the idea. Remy manages the project.
For most people experimenting with AI video, yes, at least as a starting point. The Blender-to-AI-video pipeline can produce more polished, fully custom 3D renders, but it demands real 3D skill and patience with render times. A comparison of a classic film’s opening camera move, recreated in Blender and then rendered out, illustrated the tradeoff well: the camera move got close to the original but wasn’t a perfect match, there was drift in timing, and the render itself took a long time to complete.
Cozy Clay trades some of that fidelity and control for speed and accessibility. It’s not going to replace a full Blender pipeline for someone doing serious 3D production work, but as a fast way to lock a camera angle, test blocking, or hand an AI video model a clearer spatial reference than a text prompt alone, it fills a real gap. Being browser-based and free removes the two biggest barriers (cost and hardware) that keep a lot of people from experimenting with previs at all.
How does this fit into a broader AI video pipeline?
The realistic workflow looks like this: build a rough scene in Cozy Clay, lock your camera angle and character blocking, export a frame or sequence as a reference, then feed that into your AI video model of choice (Kling, Seedance, Minimax, or whatever you’re using) as an image or motion reference rather than relying on text description alone. The previz stage isn’t meant to be the final output. It’s scaffolding that gets you closer to the specific shot you have in your head before you spend generation credits or compute time on the real thing.
This also matters more as video models keep improving at motion and texture but remain inconsistent at spatial control. Tools that let you plan a shot in 3D space, cheaply and quickly, are likely to become a more standard part of AI filmmaking workflows rather than a niche add-on.
Frequently Asked Questions
Is Cozy Clay free to use?
Yes. Cozy Clay is free and open-source, and running it through Pinocchio (also free) keeps the entire setup cost-free.
Do I need a powerful GPU to run Cozy Clay?
No. It runs in a web browser once installed, so it doesn’t require significant local hardware the way a full Blender render pipeline would.
What is Pinocchio and why use it with Cozy Clay?
Pinocchio is a free local app that simplifies downloading and running open-source AI models and tools, including video, audio, and image generation tools. It’s used as the easiest way to get Cozy Clay installed and running without manual setup.
Can I automate scene building in Cozy Clay instead of doing it manually?
Yes. Cozy Clay supports connecting an AI agent (such as OpenAI’s Codex via an API key or ChatGPT subscription, or Anthropic’s Claude) to build out scenes from natural language instructions, which you can then fine-tune manually.
Does Cozy Clay replace Blender for AI video previs?
Not entirely. Blender offers more control and fidelity for custom 3D work but takes longer and requires more skill. Cozy Clay is faster and more accessible for quick previs passes, making it a complementary tool rather than a full replacement.
