Rendering Vision with ChatGPT

Context

I began with research to find a direction for the design.

I brainstormed themes, features and materials with ChatGPT. The responses were insightful, but I found it harder to make decisions when so much synthesised information arrived instantly.

When I do my own research, I immerse myself in the sources, understand the details and slowly build context. That gives me time to connect the dots and form a story. With AI, it felt different.

It was like striking gold without having to dig—exciting, but also a little overwhelming.

The information was valuable. I simply needed to get used to this faster way of working.

Initial explorations

My first direction was to make something distinctly Indian. I explored forms inspired by Indian jali patterns and the materials traditionally used to make them.

I also wanted to solve a user-experience problem, but every improvement I considered added complexity. My instinct was to keep the object simple.

None of the directions felt right. Then, while scrolling through Pinterest, I found a cement-cast vase whose restrained material and geometry suggested another path.

Finding the design direction

One evening, while walking, I remembered the game Monument Valley. Its world brought me back to M. C. Escher’s architectural illusions, which had inspired the game in the first place.

What if a cement-cast vase could hold an Escher-like optical illusion in its architecture?

Designing with ChatGPT

I decided to use ChatGPT to design the vase. The first results were unsatisfactory. The tool seemed to lose track of the conversation, so I began to build its context deliberately, one idea at a time.

1. 01

   Establish M. C. Escher and the principles behind his work.

2. 02

   Focus on Escher’s impossible architecture and spatial illusions.

3. 03

   Connect those ideas to the visual language of Monument Valley.

4. 04

   Add the material constraints of cement casting.

5. 05

   Finally, bring the contexts together as a flower vase.

With that context in place, I asked it to design the vase. It produced several directions, but the results showed that it did not fully understand the physics or practicality of the object.

Taking control with sketches and CAD

Once the direction was clear, I returned to the tools that gave me precise control. I sketched compositions, modelled the vase myself and refined it step by step until the proportions and architecture felt resolved. Then I locked the CAD model.

Visualisation using ChatGPT

I initially planned to follow my conventional workflow: render in KeyShot and test its AI features. But getting the cement material right became difficult.

I tried a custom procedural material and one from AmbientCG, but ran into UV-orientation problems in KeyShot. Because the CAD was built in Plasticity, correcting the geometry meant unwrapping and remapping many surfaces. The process was becoming more complicated than the result needed to be.

That is when I shifted to AI rendering. I tried Vizcom first, but reaching the result I wanted would have required a custom palette and more preparation. Then I returned to ChatGPT.

Each attempt added a useful reference to the conversation. Once that library was established, I uploaded the model references and asked for realistic renders with a specific lighting setup. With a few more adjustments, the results came close to my vision.