3D software has always felt a little scary to me. Blender is powerful, but learning it feels like climbing a cliff. So when I came across this post from an AI professional, I stopped scrolling right away.
The author teamed up with architect June Chow to build a full building model in Blender. The twist is that neither of them touched Blender once. Their AI tools talked to the software the whole time.
🏗️ The challenge: 3D without the 3D skills
Building a 3D model usually means hours of clicking, dragging and memorizing shortcuts. That’s a big barrier for designers who want to explore ideas fast. The original poster sums up the payoff: “A 3D model = unlimited exploration & storytelling angles.” But first you have to get the model built.
🤖 The setup: two humans, several AI models
The pair didn’t drive Blender by hand. They prompted their way through the whole process instead. They used different methods across Claude and ChatGPT, and the AI tools sent instructions to Blender directly at every step.
Here’s what powered the project:
- GPT-6 Astra: one of the two key models doing the heavy lifting
- Opus 5.5: the other main model in the mix
- Codex: brought in later to help add a simulation
- Human judgement and feedback: essential throughout, according to the creator
That last point matters. The AI didn’t run on autopilot. June and the author kept reviewing the output and giving feedback the whole way.
✏️ The smartest trick: annotate your screenshots
This was my favorite tip. A lot of the time, the team marked up screenshots by hand instead of describing fixes in text. Why? Because visual feedback is more accurate than words.
Think about it. Typing “move the left window up a bit” leaves a lot of room for guessing. A screenshot with a circle and an arrow doesn’t. If you work with any AI that can read images, this habit can save you a ton of back-and-forth.
💧 Going further: simulating the water system
Once the building existed, the post’s author got curious about how its water system works, from rainwater harvesting to recirculation. After a chat with June and Codex, they added a simulation right into the model.
They’re not stopping there, either. The next move is an explainer video built from the same model.
📊 The result: the hard part comes first
The creator is honest about where the effort went. The hardest part was still building the first 3D model. That’s where they went back and forth the most.
After that, everything sped up. Refining details, adding a new feature or trying a new storytelling angle was often just a prompt away.
Why it matters: the first model is the investment. Once it exists, every new idea costs a prompt instead of a weekend.
🛠️ How to try this yourself
- Pick a 3D tool your AI can connect to, like Blender.
- Describe the basic structure clearly, and expect to rework this first version a lot.
- Take screenshots and mark them up with arrows and notes whenever something looks off.
- Once the base model is solid, add features like simulations or new camera angles one prompt at a time.
- Stay involved. Your judgement is what keeps the output on track.
🚀 Why I’m excited about this
According to this industry pro, the latest LLMs, with Astra and Opus 5.5 leading the pack, have kept surprising them over the past month. I believe it. Not long ago, “AI that runs Blender for you” sounded like a flashy demo. Now it’s a working building model with a water simulation.
The author’s advice is refreshingly simple. Not sure whether an idea can be brought to life? Just pick the right AI model and ask. You might be surprised.
Check out the full LinkedIn post to see the model and the water simulation for yourself.