Andrii Berezin told us about the workflow behind the Headstone Angel project, explained how the textures were edited, and talked about setting up lighting to create a dramatic look.
Hi everyone. My name is Andrii Berezin and I am from Ukraine. I have been in the CG industry for more than 5 years and I can say that I found my calling in this field. I started my route with photogrammetry, then for me as a beginner, it was important to get a primary quick result. The scans gave me that. This process of creating 3D models gives the maximum level of realism by which I am personally delighted.
When all this began, there were very few tutorials and it was necessary to collect information bit by bit. During my studies, I met wonderful guys who helped me, I am very glad that we are friends now.
I have participated in several projects dedicated to the preservation of the national cultural and historical heritage of Ukraine. It was a voluntary activity that gave me very necessary experience. After that, I began cooperation with international companies, in which I was getting specialized in photorealistic digital human photogrammetry reconstruction and creation, optimized 3D game engine-ready prop and character assets. Due to the terms of the contracts, I was not allowed to add the projects I was working on to my portfolio. So I had to find time for personal projects and development as an artist.
The Headstone Angel Project
A few years ago, I visited Lviv. My aim was to get more knowledge about local sculptors creating in directions of late romanticism and realism. A huge number of ancient sculptures inspiring me were presented in the historical and cultural museum-reserve of national importance "Lychakiv Cemetery". The cemetery was opened in 1786. It is one of the oldest existing cemeteries in Europe and Ukraine.
The angel sculpture stood far away from the hiking trails, I noticed it accidentally and could not pass by. The shapes, cuts, and finishes of the stone, as well as the dramatic story skillfully conveyed in sculpture by the famous sculptor Julian Markowski (1846-1903) could not leave me indifferent to this diamond.
The sculpture was overgrown with grass and other vegetation, and this greatly interfered with the shooting conditions. Moreover, a variety of curly plants passed directly up the stone, and in some places, weeds grew from crevices and depressions. Of course, this all a bit complicated the process of obtaining a high-quality scan. I remember I took a broom, forceps, and other necessary tools and spent several active hours carefully tidying up the monument and the territory around.
After cleaning the monument, I got to work using a Canon 750D SLR camera with 24mm and 50mm lenses. Photo scanning took place in the open sky with dynamic lighting changes, I had to change the camera settings frequently. It was important to take photos of the object in scattered light in order to minimize the number of shadows, especially to exclude sharp shadows and, accordingly, their display on the diffuse. The sun was hiding behind clouds, and at this time, I was in a hurry to take as many photos as possible.
Due to the haste and other inconveniences of the land associated with the terrain, I got a lot of blurry pictures. This resulted in 534 photos in total, 439 of them of acceptable quality that I selected in Lightroom. I always work with RAW images to maximize the control and quality customization of Exposure, Contrast, Highlights (down), and Shadows (up).
From my experience with photogrammetry software, Agisoft Metashape gives the best diffuse map result and RealityCapture gives a good mesh quality. But taking into account the characteristics of the scan object, I decided to limit myself to using only Agisoft Metashape.
Above, you can see the results of aligned photos in Agisoft Metashape – point cloud from the data. Blue rectangles are the places from which the photos were taken. Here, you can observe that the shooting conditions were influenced by trees, fences, and tombstones around.
After the Build Mesh process, we can see the result of the reconstruction. Due to the lack of information, some places in the geometry of the object were left not demonstrated. There are several reasons: features of the geometry of the object and the absence of shots from height.
We turn to clean up in ZBrush. The overall result seems to be good enough but not ideal. Preparing the model for retopology, I removed the extra parts of the mesh, used Close Holes, and, for further convenience with the topology, used DynaMesh, reducing the Active Points of the model from 10M to 200K.
To get an optimized low model, the retopology process in manual mode would take about 10-12 hours, according to my estimate, and I decided that it would be right to save this time and change the topology automatically. So, I used ZRemesher for automatic retopology and used Polypaint to influence mesh density in the right places of the model.
After the final correction of the received low model topology and the elimination of mesh defects, I began the process of cleaning the mesh. I got 5SDiv levels and Project geometry from the scan to each of them. It may also be useful to enable Morph Target before this procedure.
It helped me to fix artifacts after Project. The smooth brush also does an excellent job in this particular situation. There are many similar problems throughout the entire geometry, some are quickly eliminated and others have to be tinkered with. The process of cleaning the sculpt with the sculpting tools was long, I worked hard at every SDiv level.
To do this, I collected specific support material in PureRef, which significantly helped me to feel the shapes and restore them. Looking at the references made the work easier and clearer.
Model cleanup results
The sculpture forms had to be restored and edited. Most attention should have been paid to the face, toes, and fingers, as well as drapery. To achieve a high-quality result, I worked out each fold of clothes from the first SDiv levels to the last.
Next, I used RizomUV to create the UV and moved to Build Texture in Agisoft Metashape. I should add that for a better generation of the diffuse map, it is worth using a high poly model. But even in this case, the result will require edits and corrections.
There are a lot of artifacts, in many places, there is a blur of textures. I generated a 16K diffuse map, and, to achieve high quality, used Mari tools.
At this stage, I do the main work on editing textures in Mari with the Clone Stamp tool and also frequently use the ability to project some areas from the photos to the problem areas of the texture. It may be quite enough to get a good result in the association of these two tools.
It is also crucial to pay attention to the shadows and darkened areas of the texture. They had to be brightened as much as possible. It is now easier to create the rest of the textures for the model using the already good Diffuse texture.
For example, if you need a High Pass texture to get fine detail and the high poly model didn’t look overly polished and unnaturally clean, this texture is easy enough to get in Photoshop.
Continuing using ZBrush and importing the resulted texture as a Displacement map, I get the desired detail for the model. It was necessary to use Layers and Morph Target again to better control the application of the map. Now, the high poly model can be considered fully completed. And, after baking the Normal map, Cavity map, and AO, I move on to the render.
Above, I tried to show the stages of the lighting setup and the types of sources of light in the Marmoset Toolbag that were used for the final render. First of all, I decided on the main source of light and its intensity and then moved on to choosing the right angle. The choice of secondary filling lights and their types required constant experimentation. I tried many options and consulted with friends each time setting up the lighting differently.
The drama of the sculpture was disposed to general contrasting lighting, but I decided to play precisely on the contrast of warm and cold tones. I tried to achieve the effect of bright lunar mystical lighting by getting the balance between many different types of sources and their settings.
At the same time, I tried to make the shaded areas of the image more readable. The right light solutions are very important to me in the rendering process. They can make the final result spectacular and interesting by focusing the viewer's attention on the main thing in the object.
I tried to make the most of the capabilities of Marmoset Toolbag for minimal post-production and I hope I managed to give some interesting color effects to this final render.
That was my workflow on this project. There were no main challenges here because I knew what I was doing well. But as a beginner, I would probably have encountered difficulties ranging from photography skills and special settings for the photo equipment to the correct lighting setting for the render. To make the results of photogrammetry to be of acceptable quality, I tried to delve into each stage of the pipeline and conducted many different experiments. But in general, everything consists of simple and even elementary things, it's just that there are a lot of them in the beginning.
I hope you’ve found something interesting and useful in this article. Thank you for reading and I wish you to have good weather for photogrammetry, guys!
You may find these articles interesting