Realistic Character Workflow
This final chapter is more like a small post-mortem. There were many things that could've been done better during the creation of Einar in Project Heist. It's good to document and share our notes we want to address the next time and unclear questions we have.
While the concept of shapes worked well for us in theory and provided the most amount of freedom for each deformation, it was ambitious to do this manually instead of relying on scan data.
An ABC wrinkle map workflow would've suited as just fine and could've made certain transitions between facial deformations more smooth. Iteration and changes would've also been faster to implement. Focusing on just three high frequency wrinkle maps for both bump and displacement would've also given us more time to get a higher level of detail and realism.
That doesn't mean the shapes workflow is not something we'd peruse in the future. But this workflow would be most ideal if we'd use scan data as a base or work on improving the Multires sculpting support in Blender. It would make the process easier and the results more strikingly realistic.
Early on we decided to make our main character a ~60 year old man. The reasoning was that the amount of wrinkles would make the realism easier. You'll focus much more on the details on a young and smooth face after all.
But this was slightly misguided since executing realistic skin deformations on an older character is just as challenging, if not more.
This is again where scan data could've been more helpful to get the age right. We relied on a lot of reference but this still cost a lot of time for creating the character and expressions.
There were many discussions and open questions on what we can or should accomplish proceduraly.
The shapes workflow for deformations and ABC wrinkle map workflow are very hand sculpted and painted. Meanwhile we also implemented dynamic wrinkle generation and mixing via tension maps and flow maps.
In the end we applied both on the character. Each added technique and adjustment pushed the realism a bit further. But now that we defined these manual and procedural workflows it would be wise to define which should be used where, for the next project or final tutorials. This should lead to a more polished workflow.
Figuring out the sculpting and baking workflow with unfinished features was challenging. It was a constant journey of discovery and adjustments.
We should've spend more time to automate the repetitive tasks of this workflows, like baking and exporting the displacement maps and shape keys.
Also switching between and syncing up the sculpt layers (high res source sculpt) and shape key sculpting (low res preview objects) could've been automated.
Because of time and budget constraints we had to cut a lot of corners. Rendering limitations also played into this. That's why the methods that are explained here could be further optimised and adjusted to match the quality and efficiency of industry standard workflows.
Further R&D and consultation with outside studios and artists will also help in achieving this. Any updates on that would be shared here as well.
Once we get back to these workflows we either need active Blender development or scripting to make the process more smooth.
Certain features are still not fully supported or face performance issues. Multi-resolution sculpting is a primary case of this. But also baking of displacement maps is still not reliable.
For realistic characters, or even sculpting in general, it's also a high goal to add the feature of layered sculpting officially. But this will heavily rely on a rewrite of the Multires sculpting. Many issues in the workflow are rooted in this.