Rising up in a metropolis neighboring Lyon, France, I’ve been lucky to witness the celebrated “Fête des Lumières” firsthand. This fascinating annual event bathes Lyon in an beautiful showcase of sunshine, metamorphosing its thoroughfares, iconic websites, and edifices into awe-inspiring masterpieces. Annually, tens of millions of holiday makers from across the globe flock to witness this extraordinary spectacle, immersing themselves within the enchanting realm of sunshine, colour, and innovation that defines the Fête des Lumières.
At present, I made a decision to embark on a private problem: utilizing Houdini, a 3D software program, to create and simulate a lightweight artwork set up paying homage to these seen on the “Fête des Lumières”. This concept has been brewing in my thoughts for fairly a while; I’ve envisioned a novel strategy. I aimed to position the customer on the middle of the expertise by incorporating an onsite immediate system. This method would dynamically generate the projected picture over the constructing primarily based on prompts supplied by guests, using generative AI expertise. With that match plan in thoughts, let’s see how we will create a proof of idea that does this.
In a earlier article, I demonstrated using SideFx Houdini software program (sidefx.com) to generate AI-generated pictures by way of the MLOPs bundle (github.com/mlops). You’ll be able to take a look into the detailed rationalization within the earlier article (here) to understand the basics of my Steady Diffusion picture era workflow. On this article, I’ll delve deeper into extra elements of the method, constructing upon the groundwork laid out beforehand.
First issues first, let’s discovered a 3D mannequin value of our inspiration. On my facet, I’ve settled on utilizing a 3D mannequin of the door from the French medieval church of Sainte-Radegonde (Poitiers). The mannequin is freely accessible for obtain (here).
Now that we’ve downloaded and imported our mannequin into Houdini, we’ll apply a collection of transformations to align it with the viewport grid. We then cut up the mesh into Houdini teams, which we’ll later use to populate a customized attribute named “building_parts”. This attribute will show helpful shortly once we delve into the generative AI step, as it is going to be used to create masks I’ll use to merge all our generated AI layers utilizing compositing.
For this proof of idea, we’ve restricted ourselves to a few zones of curiosity: the primary door group, encompassing all areas inside the arch; the sidewalls, consisting of the 2 facet columns and the primary wall; primarily all the pieces else. At that time, our setup that handles the mesh appears to be like just like the one under.
