Inspiration
I had recently learned Blender and was making 3D Objects in it, I had heard of sites online that claimed to be able to generate 3D objects using prompts but these were often either not as good as advertised, far too high poly to be practical or locked behind an expensive paywall. I had wondered if an accessible AI such as Gemini was capable of the task and began to experiment.
What it does
It takes a prompt from the user such as "A wooden chair" or "A great staff" as well as their API key they can get from Google AI and then generates a 3D model based on that, however what it is actually doing is generating Three.js code that will result in the 3D model the user desires. It then allows the user to view the object on the web in a 3D interface and provide the object to download it as a .glb file which can then be used in other tools.
How we built it
The code itself is quite simple, it is a static html web page with JavaScript included that simply sends a detailed prompt to Gemini, using a description from the user and a provided API key. To speed up the process I was able to use Gemini itself to help generate the code and styling for the site to work and look good as well as plan the development of further features.
Challenges we ran into
The main challenge is when generating more complex objects the model is not as successful at creating them due to how much more detail is required.
Accomplishments that we're proud of
The site works and is very easy to use, anyone can click onto the site send a prompt with their API key and get a 3D model they can download and use within seconds. This is very useful tool that I will likely be using myself for getting a jumpstart when creating 3D models in tools like Blender.
What we learned
I was unaware just how effective creating 3D objects using something like Gemini would be and now know just how much potential this technology has when it comes to creating.
What's next for 3D Object Generator
To push its limits further and see is it possible to create more complex objects or environments using the tool by using more detailed prompts.
Log in or sign up for Devpost to join the conversation.