Using Blender with ChatGPT

Puter demonstrated building a 3D ice cream cone model in Blender directly from a text prompt using OpenAI's GPT-6 Astra model driving the browser-based Blender through ChatGPT's Computer Use feature, according to the company's writeup. The approach runs Blender compiled to WebAssembly with a WebGPU backend in a Chrome tab, letting the ChatGPT browser extension click menus and type values instead of generating Python scripts. Puter cited GPT-6 Astra's reported 72.6% score on OSWorld 2.0 versus 65.7% for GPT-5.6 Sol as the reason for choosing the model.

Using Blender with ChatGPT: Build 3D Models from a Prompt In this article, we'll show you how you can use Blender with ChatGPT to build and design 3D models, compare the methods available today, and see which one works best for you. How ChatGPT can work with Blender There are several ways to use Blender with ChatGPT, or with any AI model. The most common one you will find is having ChatGPT write a Blender script in Python, which you paste into Blender's Scripting workspace and run. A more elegant version of the same idea is Blender MCP https://github.com/ahujasid/blender-mcp , where an add-on and a small server connect the AI to a running Blender so it can run code and read back the scene on its own. Most AI integrations with Blender work this way. You supply Blender with code and Blender builds it. This is good for procedural objects like tables, rooms, and product shots. The downside is that Blender's Python API changes between releases, so generated scripts often break on deprecated calls, and with pasted scripts the model never sees the result. MCP fixes the feedback loop but needs Blender installed plus some setup, and its ChatGPT support is still evolving https://blendermcp.org/setup/chatgpt . AI models have since gotten better at using a computer, and that opens a third way. Instead of talking to Blender through Python, ChatGPT can look at the screen, click through menus, and type into fields, the same way you would. You let it operate your computer and it designs the model in Blender's own interface. Using Blender with ChatGPT in the browser We do this in the browser, with our Blender in the browser https://developer.puter.com/labs/blender-web/ and the ChatGPT browser extension https://learn.chatgpt.com/docs/chrome-extension . We compiled Blender to WebAssembly with a WebGPU backend, so it is the real Blender interface running in a browser tab source on GitHub https://github.com/HeyPuter/blender-wasm . Because it is a web page, the ChatGPT extension can drive it like any other site. This lets you go from a prompt to a 3D model in a few minutes without installing Blender. This assumes you already have the ChatGPT desktop app installed. 1. In the ChatGPT desktop app, open Settings Computer Use , pick Chrome, and install the extension. 2. In Chrome, open Blender in the browser https://developer.puter.com/labs/blender-web/ and click Launch Blender . It needs WebGPU, so use a recent Chrome or Edge. 3. Back in ChatGPT, pick GPT-6 Astra from the model selector, and describe what to build. We used GPT-6 Astra /ai/openai/gpt-6-astra/ because it is the latest model and OpenAI built it around computer use, with a reported 72.6% on OSWorld 2.0 against 65.7% for GPT-5.6 Sol. We covered that in our GPT-6 Astra review /blog/gpt-6-astra-review/ . Our prompt was: make me an ice cream on a waffle cone with cherry, sprinkles, and chocolate stick only use shapes, no commands or scripts, and no need to save the scene GPT-6 Astra building the scene in Blender in the browser. Sped up 11x. It takes a screenshot, decides on an action, clicks or types, and takes another screenshot. It used the Add menu to create primitives, the N panel to set dimensions and positions, and the Material tab to add and color a material. When it misplaced an object it noticed on the next screenshot and moved it. The result worked well. It built out everything we asked for, positioned correctly, with the material applied. It is slower than running a script, since every step is a screenshot and a click, but for a scene like this we found that an acceptable trade for not writing or debugging any code. Which one should you use We built the browser version of Blender, so that is the approach we are going with. If you do not have Blender installed and want to try this quickly, use the browser version with the ChatGPT extension. There is nothing to configure and nothing to debug. If you already have Blender installed and your workflow lives there, do it there. Blender MCP gives the model a faster, richer connection to your scene than screenshots do, and the setup is a one-time cost. Conclusion You can use Blender with ChatGPT by opening the browser version of Blender and letting ChatGPT, through the browser extension, interact with it directly. It works from the interface rather than from generated code, so there are no scripts to paste and no API calls to debug. Try it yourself at developer.puter.com/labs/blender-web https://developer.puter.com/labs/blender-web/ . Related Ship a Full-Stack App with One Prompt Create a to-do list app using Puter.js Coding manually? see the guide https://docs.puter.com/getting-started/