This project was part of a series focused on creating high-fidelity approximations of monumental Paris landmarks.

The Palais Garnier presented a particularly demanding 3D environmental modeling and texturing challenge because of its scale, architectural density, ornamental detail, and large number of facade sculptures.

The goal was to create a recognizable, game-ready approximation rather than an exact replica, while meeting the technical requirements for integration into GTA V.

LOD0: approximately 900k triangles

LOD1: approximately 25k triangles

Separate decal model covering the LOD0

Geometry fitted to the supplied ground and road layout

Diffuse map

DirectX Normal map

Specular map

Maximum texture resolution of 1K

Reasonable material count relative to asset complexity

The building was created as a large environment asset with dedicated LODs and a separate decal mesh. The high-detail LOD0 preserved the architectural character of the Palais Garnier, while the significantly reduced LOD1 maintains the overall silhouette and readability at distance.

With 900k triangles at LOD0, the reduction to approximately 25k triangles for LOD1 is a significant optimization step.

Blender

Substance 3D Designer

Photoshop

ChatGPT

ComfyUI

I decided that, for most of the building, the best approach is to build it with modular pieces. This allowed me to efficiently assemble the facade while maintaining consistent proportions, topology, and detailing.

The textures are: Diffuse (opacity in the alpha channel), Normal (DirectX) and Specular with a maximum texture resolution of 1k. They were created combining existing image assets, Substance Designer texturing and AI image upscaling. I have used Photoshop heavily in this project for texture preparation and assembling the atlases.

One of the biggest challenges was the sheer number of architectural sculptures and ornamental elements required to reproduce the Palais Garnier.

I used an AI-assisted workflow to accelerate the texture production pipeline. In order to create the inordinate amount of sculptures and textures required to make this monument look amazing, I used ChatGPT Go and fed it the reference images I collected in order to get cleaned-up, upscaled and adjusted versions of that reference I could use as a texture. The texture processing quality was excellent and this project would have been impossible to finish within the given time frame without this workflow.

Another AI-assisted workflow was generating 3D statue models from the statue reference images. I don't have much experience with this yet so the results were hit and miss, but most of the statues in the final 3D model were ultimately generated with AI, with some manual clean-up afterwards. The quality of the generation greatly depends on the starting reference images, which were not perfect. The entire workflow for the sculpture atlases and 3D sculptures was as follows:

Create materials in Substance Designer/Use existing texture sets from my PC/Internet

Use the base colors as a background images in Photoshop were I assembled, cropped and unified the visuals of the AI-generated sculpture images as part of a dense texture atlas (the Sculpture XX textures you can see in the post)

Make masks for different materials on the same atlas

Import the sculpture atlas textures and masks back into SD, generate a normal and roughness/spec map, add damages and details, adjust everything to make it look natural

Export the finished Diffuse, Normal DX and Specular maps as .png images

Generate the statue 3D model from a real reference/AI enhanced reference

Optimize and fix the generated 3D model

Project from front view the statue texture on the 3D model

This workflow is really incredibly quick for generating decent 3D statue models with the texture projected from the front. Of course, this can only work on the sculptures are part of a facade and you can't go behind it since the project technique doesn't work on the back, and the sides are also affected.

In the end, I had a wonderful time working on this technically innovative project (for me) since I love learning new stuff and making awesome 3D environment art :)

I hope you had a great time reading this as well!