Making My First CGI Character from Photography
Think Tank Training Centre student, Pedro Costa, embarks on his inaugural journey into creating a CGI character and generously shares valuable tips and insights acquired throughout the process.
In the second term of his studies at Think Tank Training Center Online, Pedro Costa successfully completed his inaugural CG character project. Over a span of 5 weeks, Pedro took on full responsibility for all aspects of the project, with the valuable guidance of a class supervisor provided on a weekly basis.
When tackling the task of creating a CG likeness of a human character, it is a common practice for both beginners and professionals to choose famous actors, actresses, or celebrities as their subjects. However, this approach comes with several challenges, especially for beginners. One major hurdle is the scarcity of high-quality, close-up images available on the internet. Moreover, the lack of photo metadata further complicates the process.
Having acquired extensive experience as a 2D character artist, I embarked on this project with a strong sense of confidence. However, my journey was riddled with numerous challenges along the way.
In this account, I will delve into the process and concepts underlying my choices, as well as highlight the multitude of poor decisions I made. By sharing my experiences, I hope to assist others in saving valuable time when undertaking what is arguably one of the most demanding tasks in the realm of 3D: the creation of a photorealistic human character.
As a devoted fan of Marvel, and particularly of Spider-Man, I made the decision to embark on the challenging endeavor of creating a film-quality rendition of the Iron-Spider suit worn by Tom Holland. This iteration of Spider-Man holds a special place in my heart as my personal favorite, and I was excited to tackle the unique design of the iron spider suit, which sets it apart from the classic versions. Opting for something new and different proved to be an excellent choice for my creative exploration.
Collecting reference images
The project's greatest challenge lay in capturing the likeness of the character. As it was my first time attempting such a task, maintaining an accurate representation of the individual's features while ensuring correct anatomical proportions proved to be particularly demanding. Striking the right balance between these two elements posed an ongoing struggle throughout the project.
When tackling a project like this, make sure to do the following:
1. Gather high-resolution images whenever possible. While it may not be mandatory, higher resolution images provide better visibility of intricate details.
2. Select reference images from multiple angles. Having a variety of angles helps to capture a comprehensive understanding of the subject.
3.Consider the lighting conditions in the reference images. Shadows can offer valuable cues for determining the shape and contours of specific areas.
4. Utilize software like PureRef to consolidate and organize all your reference images in one convenient location.
5. Categorize your reference images for easy navigation when required. Sorting them by relevant categories streamlines the workflow and facilitates efficient access.
6. Prioritize images captured using longer focal length lenses (85mm and above). For instance, red carpet photographs often serve as excellent references. Shorter focal lengths can introduce distortion, making it more challenging to accurately match facial features. However, close-up pictures can still be beneficial for observing skin details.
By following these guidelines, you can optimize your reference collection process and enhance the accuracy and quality of your work.
As a novice in this field, I adopted a progressive approach to enhance my learning. Beginning with a simple sphere, I gradually advanced to sculpting a skull and subsequently incorporated the skin layer to attain the desired facial anatomy resembling Tom Holland's. In my experience, commencing with a skull foundation proved to be more advantageous, enabling me to grasp the intricacies of facial proportions more effectively or you can start with a base mesh with clean topology.
It is crucial to maintain subdivision levels 1-2 during this stage of the process, as these levels establish the primary forms and even some secondary forms. Personally, I found this to be the most demanding aspect, despite having prior experience with anatomy and years of 2D drawing practice.
In Zbrush, there are several methods to utilize reference images effectively. One of the most versatile and preferred approaches is using Spotlight. Additionally, you can also consider using image planes in the Texture Palette or the Draw Palette, as well as employing the See-Through method.
After achieving a basic likeness, it is crucial to avoid becoming too fixated on that stage. It's important to recognize that sculpting corrections and adjustments will be an ongoing process throughout the project. As you incorporate elements such as Subsurface Scattering (SSS), grooming, and physically accurate lighting, your perception of the sculpt may change from what you see within Zbrush. This iterative back-and-forth workflow can be challenging at times, but it is an integral part of the overall creative process.
To tackle the hard surface components, specifically the suit, I initially utilized dynamesh in Zbrush to establish the overall shapes. This approach provided the advantage of swiftly blocking out the desired forms without being concerned about topology. Subsequently, I transitioned to Maya to perform retopology, ensuring a refined and clean topology for the suit.
Wrapping head to a good topology
Given that I am utilizing the V-Face Texturingxyz pack from the Texturingxyz website, which includes a base mesh with a well-structured topology, I leveraged this asset in R3ds Wrap to transfer my dynamesh sculpt onto the base mesh. This process enabled me to seamlessly transition from the initial dynamesh sculpt to a base mesh with optimal topology.
For guidance on using R3ds Wrap, you can find tutorials available here. It's worth noting that if you are a student, R3ds offers a complimentary student license (accessible through the "Buy" section and selecting "Student").
Skin Detail Phase
This can be done in many ways, but the more or less “industry standard” method is taking scan based data (Texturing xyz, 3D Scan Store are the best options) and either projecting or wrapping this to your model. In my case, I purchased the VFace from the XYZ store.
For the texture transfer, i used Mari but you can still use the popular method using the R3DS Wrap. You can find this workflow in Texturingxyz Youtube Channel, here.
After this process is complete, you should have three transferred texture images: Albedo color, packed RGB displacement, and packed Utility map.
Following that, it was necessary to refine the displacement using projection techniques in Mari. Subsequently, the data was split into three separate maps using the Copy Channel node, with XYZ serving to differentiate between secondary, tertiary, and micro details. The advantage of working with Mari is its compatibility with 32-bit textures, eliminating the need to compress the data and preserving its quality.
Please note that there seems to be some confusion surrounding Zbrush, displacement bit depth, and .exr files. Here are the facts: Zbrush can handle 32-bit .exr files, but it cannot process negative values. Therefore, when importing a displacement with a midpoint of zero, the midpoint must be adjusted for proper functionality.
Moving forward, each displacement channel was imported into Zbrush as its own layer. This approach allows for finer adjustments and greater control over the intensity of each channel. Achieving a balanced composition in this regard is crucial. Following this step, additional layers of skin detailing were added to address any blank areas and modify wrinkles and creases as required by the character's expression. This process may be time-consuming, particularly for those who are less experienced in executing these intricate adjustments.
I divided texturing into two parts: The skin and eyes would be textured in Mari at 8K, and everything else (suit) would be done at 4K in Substance 3D Painter.
For the skin, I started with the XYZ albedo map, as well as some of the low resolution skin textures that came with my base mesh. Lots of touchup, color grading, and removing shadows was the primary work done here.
For the suit, Substance 3D Painter made the texturing process easier. The metal look and patterns required only small adjustments and tweaks to look better. This was probably the easiest part of the project.
As I initially modeled the suit in an A-Pose, while my final shot required a more neutral pose, I employed Maya and the Blend Shape tool to minimize distortions as much as possible. By utilizing this technique, I was able to achieve a seamless transition and maintain the desired aesthetics of the suit in the final pose.
Mastering the art of grooming in Xgen is a delicate and intricate process, requiring a solid foundation of essential fundamentals for achieving success:
1. Create a "scalp" mesh. UVs must be in the 0-1 space and with lambert material applied.
2. Keep a second “utility” set of UVs matching the original model for updating the shape. This is very important, you need to be able to update your groom shape as the sculpt evolves.
3. High resolution reference for all angles of the groom.
4. Focus on guide work, the guides are the important part cause they will give the proper flow and direction of the hair. If the shape and flow aren’t correct, no amount of modifiers will fix it.
In my personal experience, I undertook the entire grooming process for the face, meticulously working on everything from the hair to the peach fuzz. By utilizing guides, I generated the initial clumping map and further enhanced the hair by incorporating at least two additional Clumping modifiers. With the Clumping successfully implemented, I proceeded to apply Noise and Cut modifiers to introduce texture variations to the hair. Additionally, I explored various other modifiers to ensure the best possible outcome for my character, adapting as necessary to achieve the desired result.
Look Development/Rendering Phase
For this particular project, I made the decision to transition to ACEScg as my working colorspace. There were several compelling reasons behind this choice, with the primary factor being that ACES has become the industry standard within Film pipelines. By adopting ACEScg, I aimed to align my workflow with current industry practices and ensure seamless compatibility with other professionals in the field.
The technical aspect can be intimidating, but there is great information available on ACEScg, namely here.
Mastering a realistic skin shader presented its challenges, but with the aid of excellent tutorials like Adam O'Donnell's Obi-Wan Kenobi tutorial series, I managed to achieve a satisfactory outcome.
The key approach is, to begin with a basic shader and gradually incorporate the texture maps one by one, ensuring that each map is functioning correctly before moving on to the next.
Furthermore, it is crucial to calibrate the color and exposure levels by creating a neutral environment that is free from light contamination. This allows for a clear assessment of the texture's appearance without the influence of colored light from an HDRI. Incorporating an ACEScg Macbeth Chart in the scene facilitates the correct balancing of lights and prevents overexposure.
Coloured light and exposure can be used purposefully afterwards, but only once a good neutral base has been established.
My lighting setup was with the amazing Photo Studio Environment by CAVE Academy, designed by Jahirul Amin, where I used 3 lights, two Key lights, and one Fill light.
My intention with this article is to provide valuable insights that can assist others in overcoming the challenging learning curve associated with creating realistic human characters in CG. I encourage you not to lose hope and persevere, as the process can be incredibly rewarding in the end.
- PureRef for the organization of reference images.
- Zbrush for sculpting, modeling, and baking displacement maps.
- Maya/Arnold for modeling, retopology, UV layout, lighting & lookdev.
- R3DS Wrap for geometry wrapping.
- Mari for skin, eye & texture transfer.
- Adobe Substance 3D Painter for the suit.
- Adobe Photoshop for light compositing.