A LIFELIKE PORTRAIT: Breaking Down Photorealism

By
-
Login to Follow
-
Industry
  • Film & VFX
Subject
  • General
Products
  • Maya
  • Mudbox
Skill Level
  • Intermediate
It’s in the details.

Let your head do the hard work for you... or something like that. A few weeks ago, after seeing an illustration I created, Fianna Wong from AREA approached me, asking if I could write a tutorial about how I created the piece. I couldn't find a way to approach the idea – I didn’t feel the image had any technical merit that hadn’t already been thoroughly covered in other tutorials. It didn’t seem compelling to scrutinize stubborn details that anyone could recreate if I gave them all the files I used to create it. Upon reflection, it occurred to me that there was, in fact, something I wanted to talk about – I needed to show you my creative process. I always have been fascinated by the appearance of people; the outside part of them, their body language. The way all the muscles arrange themselves to form the expression that perfectly conveys the intent and emotion of the moment. How many fantastic people have you met just looking at photos of people online or in photography books? How many of the iconic images of Steve McCurry do you remember? What is in those memories? I see stolen emotion fixated in an instant, the reflection of a corner of your own soul. Something you know well, something both the subject and yourself share. This is empathy.

Image above shows that it's about the details. When I decided to create what became “Shades of Dambu”, I wanted to know if I could achieve a similar result by creating and sharing such an honest instant in a completely artificial manner. It is something I have always attempted in my personal work, but never to that level of detail. This task is generally the trade of the inspired photographer. Can I make a non-existent character feel alive to the eyes of the viewer? Perhaps; first, I had to review the tools at hand to be sure to use the right mix. For the goal I had in mind, I used tools that are comfortable and familiar, having used them for years in many different projects. For sculpting organic surfaces and characters in general, I chose Mudbox. I love it for its simple uncluttered interface and unconvoluted workflow. For all that is structural modeling, scene composition, framing, and previz, I use Maya. It has no dead end and is very open about everything it does in the background. Also, if you miss a function in Maya, you can create it on your own, down to the very core of the platform. That's why I love Maya: it can bend to my will to become the tool I want. You can do that as well; open the script editor, log everything Maya does as you work, learn the language a bit at a time and press F1 to refine your knowledge. Maya's documentation contains everything you need to know about Maya. I found this selection satisfying, as one of the key components of any talent is to know the tools of your trade like the tip of your fingers. Every level of experience is valid because ultimately, it is the constant process of learning and repetition that gets you where you want to go. Starting at zero, you make your first marks an experiment, always building, making sure every experiment teaches you something new or more complex. This process repeats until you gain the confidence to attempt something bigger, and then let that become your new standard. By repeating steps that bring good results, soon, your muscle memory and your brain learn the patterns and automate the technical aspect of the work for you. This is the mastery of the most important of the tools. It is in this way that we realize that whatever your choice of tool is, the same processes apply - repeat operations until you know them well enough to move on, remembering never to stagnate. Challenge yourself.

- The Uncanny Valley-

So, where do you get from software use to self-discipline to a portrait on your screen? Well, I find the only way I manage to achieve this is by letting my instincts show me what will send the message. The human brain spends its time analyzing snapshots of perception to make sense out of them, but that is a process that takes time. Faces and expressions contain information that is coded with a lot of subtlety, yet understanding a human facial expression is an almost instant process because your brain will fill in the blanks of what it has not yet captured. The vast majority of people only need 3 dots to imagine a face: two eyes and a mouth, the major emotional medium of the human being. Even more interestingly, the level of detail of your imagination depends on the level of complexity of these 3 elements. The more precise the eyes and mouth are, the sharper the imagined face becomes but it takes more time for your brain to get there. By prolonging the process, a pitfall opens up; if this extra data includes an unexpected defect, your attention is drawn to that point. This anomaly becomes a distraction and the "connection" is lost.

The image above is a demonstration of how your brain extrapolates available information to fill in the blanks. If it had to be photorealistic, then I needed to integrate a really good rendering engine into to my toolbelt. I chose a GPU-based unbiased renderer for that task. Octane render is a small, easy to learn software but it has deep possibilities that are still growing. Most importantly, it allows me to easily and steadily move towards my own vision without losing it; I use it in a repeat fashion from the beginning to the end of the process to constantly preview that my work is going in the right direction. The best of all is that Octane Render's speed of convergence allow for a near-real-time creation process. As with painting or drawing, the immediacy of the results allows for an immediate reaction. It then falls to trusting that an anomaly should be corrected, making the change, and returning to reach the next level of detail. When I started working on this character, I knew I wanted her to be photorealistic and very detailed, enough that she could push the viewer's disbelief beyond the uncanny valley, and provide “enough.” On the other hand, I also knew that I didn't want her to look completely real. I wanted my work slightly stylized, with an elastic pose, stretched proportions, emphasis on the archetypal traits... Everything tangible enough that it almost could be real but different enough that the viewer objectively knows it is not. Inspired, practiced, and with tools in hand, I began my creation process. Let me break it down like a recipe:

1- Export the head of character from Mudbox and import it in Octane. At this point, it does not have to be very detailed. If it is, even a high resolution mesh will do, you can optimize later if and when you need to save GPU memory. 2- Assign a good looking skin shader to your model. A good starting base for Octane was created and generously shared by user TonySculptor and is available in the software's material library. 3- In Mudbox, create a simple base texture for your model; loosely define the colours of the facial features in broad strokes until it feels natural. Contrasts will be faded by the shading, so do not hesitate to over-exaggerate if needs be. 4- Load the texture in Octane, link it to the skin shader and start tweaking shader and lighting values until you get something that resembles actual skin.

Image above is Step 4 AKA The inevitable disappointment. This was the first time I imported my model into Octane from Mudbox. Suffice to say it was to take a lot of tweaking and refinement before it resembled anything interesting. 5- Your loop begins now. Look at your result; your brain will alert you to anything abnormal. Pause the render when you see what is wrong.

Image above is the result of Step 5 to step 7. Wash, rinse, repeat. Ad nauseum. It goes fast when you get the trick. You'll know when it clicks. 6- Switch to Mudbox (still running in the background,) correct your model and textures according to your observations. try to keep the corrections on a large scale – is not the time to go into the details yet. Export your model. 7- Refresh the model in Octane, and restart the render. Wait a few seconds and return to step 5. 8- It is likely that you'll spend the first few hours tweaking details around the eyes and mouth. This is natural, but remember to resist the urge to work closely on the details. Keep your renders from a good distance and your sculpting as well. If your eye doesn't detect any distracting anomaly, congratulations, you moved into the uncanny valley. The way out of there is up. Save a snapshot of your render.

Image above is Step 8! I think that's pretty much when it started feeling good. I could see her beyond the boundaries of the image. 9- Import the snapshot in Mudbox as an image plane in your viewport. 10- Look at the image for a few seconds, let it sink in, let your internal analytic engine imagine the missing details. It comes naturally. 11- Now look at your sculpture and start committing the “missing” details. The word committing is key. Export your model again. 12- Refresh your model in Octane and return to step 5. 13- If you get stuck, that means your artistic eye cannot fill in more details because crucial information is missing. Go to Maya and create facial items: Eyes, teeth, tongue, eyebrows, whatever fills in the blank. Save them. 14- Import in Mudbox, tweak and detail up to something recognizable. Export the new models. 15- Refresh the model in Octane, and now you can start looking at smaller details, keeping the bigger picture in mind. And you're back in step 5! 16- When everything feels right and your brain doesn't fill in the gaps anymore even in close up, you're done. Set the render resolution to something large and go to bed.

The image above shows the progress timeline. Do you get the drift? It takes time and patience of course but it's a simple and efficient technique. If you train yourself at that game very well, you'll realize you can use that neat little trick for a lot of things other than faces. Once you reach a certain amount of realism, your brain will hallucinate any detail that is missing. Take that detail, insert it into your work, press render and compare. Repeat until you have hit your goal.

The image above shows the final high resolution mesh of the head, 20 million polygon, here showing the texture controlling the thickness of the layer of otjize (ochre paste) on the surface of her skin. I hope you’ve enjoyed my observations and notes; in addition to this tutorial, I have also been asked to create another illustration and set of tutorials. Today, I have covered the process and procedure behind making “Shades of Dambu.” In the next set, a technical challenge makes both the work and the tutorials a challenge for both you and myself. In the meantime, here is a little teaser for my next work.

Additional Links
Hugues Giboire: 3Dpeyote (Interview)Hugues Giboire
Posted By
Tags
  • Maya
  • Mudbox
  • General
2 Comments
To post a comment please login or register
Good Work!!
| 5 years ago
thanks for this great tutorial
*Save $66 per month on Autodesk's Suggested Retail Price (SRP) when purchasing 1 year term 3ds Max or Maya subscription.