NEFERTITI BUST

The project was planned to be a stress test of RealityScan’s capabilities, which eventually turned into a test of my troubleshooting skills. An unforgettable journey through the world of photogrammetry.

Optimised version (~1.5k tris) inside Unreal Engine (Lumen).

~ BREAKDOWN SECTION ~

I used the worst-case scenario, where RealityScan was fed not with high-megapixel photos from a camera, but with frames extracted from a phone video. For lighting the model, I used the phone’s flash, which more or less evened out the illumination on the model.

For aligning top and bottom parts, masks were created for the footage of the sides of the bust, and four control points were added, which ultimately helped merge components into a single model.

Although the upper and lower parts geometrically merged into a single model, due to the difference in lighting when capturing these parts, an artifact appeared in the form of a bright stripe at the bottom (fixable).

The next step was cleaning up the model in ZBrush while preserving/restoring the important parts of the silhouette.

Then I unwrapped this high-poly model in Blender and baked the albedo from the scan onto it in Substance Painter to correct the scanning artifacts. Below is the result of manual texture cleaning in Substance Painter.

When the high-poly model and the base color map were ready, the last step was creating the low-poly. I aimed for around 1.5–2k polygons, since I wanted to preserve as many facial details as possible (concentrating the topology around the lips and nose).

After retopo i got a 1.5k tris model. The poly amount can be lowered by sacrificing the overall silhouette.

The cleaned-up high-poly (with the corrected albedo) was baked onto the low-poly to generate the maps, and then the final texturing was done in Substance Painter (additional wear, extra normal details, roughness variations, etc.).

As a result of the project, I gained extensive experience working with photogrammetry and learned which aspects to pay attention to when working with RealityScan. A note to my future self on what to focus on:

1. Don’t shoot the model on a phone. By using a good camera, I will save time on restoring texture details and improve their overall quality.
2. Lighting setup. It is crucial that the object is evenly lit from all sides so that the real-world light doesn’t bake into the texture. This helps RealityScan’s algorithms correctly stitch different parts of the model together (for example, the top and bottom). It’s important to prevent real-world light from baking into the texture to avoid texture artifacts (like the strip at the bottom of the bust with different lighting that I had to fix).
3. Invest in a cross-polarization setup. Using a cross-polarization filter will eliminate unwanted light, which again helps preserve texture details and, most importantly, reduces surface noise on the model. This saves time cleaning it up later in ZBrush.

Here you can see the model and textured details up close