I suggest that, at the very least, an artist should be proficient in texturing wood, leather, and metal, as these are the most common materials encountered in future work.
Another crucial skill is achieving likeness, as 3D artists often work with other people’s concepts. The reference I chose allowed me to practice these skills, and the absence of metal was compensated by sculpting a complex dragon head to refine my likeness skills.
The fact that this is traditional oriental art personally appeals to me and Zhelong Xu’s speech at Zbrush Summit 2019 convinced me to pay attention to traditional art with a deep cultural background.
Design and References
The reference was found on an auction house site. I recommend prominent museums such as the Metropolitan Museum and auction houses as reference sources because they provide detailed photos.
I made several alterations to the original design, including changing the soundbox to a lotus flower box, adding strings for a more realistic appearance, and omitting the seal to avoid a museum-like impression.
- Marmoset Toolbag
- Adobe Photoshop
- Substance Painter
In blocking out, I relied on creased edge modeling in Maya to crudely shape the overall form.
I neglected optimal topology for sculpting since it was clear that the main part of the props, the dragon head, would require Dynamesh.
Sculpting relied on Dynameshed subtools that compromised the edges. That is why I remeshed duplicates of those subtools with clean edges, allowing me to reproject either edges or sculpt, depending on the situation.
The subtool before projection should have a Morph Target to erase undesirable elements of the projection. Having a duplicate with subdivisions is also important because, at any stage of sculpting, you can notice a serious mismatch in general forms between the reference and your work.
It is extremely challenging to correct if you have only a high-density subtool, but relatively easy if you have a subdivision with a light polycount.
As for the soundbox, the main issue was placing petals evenly and avoiding their overlapping. To achieve this, it requires the largest layer of petals to have an amount of elements that can divide 360 degrees without a remainder.
In my case, that is simply ten petals; if they occupy 36 degrees of the shape, they make a circle. I created a cylinder the same size as the soundbox with 10 sections.
Then, I blocked out and sculpted a petal from a single section; the rest was just a series of copy-and-place actions. This approach suggests that details should be sculpted last to avoid a cloned impression.
As for the floral ornament, since I didn’t have the back view, I only had partial information about the ornament.
It took many tries to complete the ornament in a cohesive manner, which is why keeping the Morph Target of a subtool as an eraser is important.
This is more convenient than tracking back in history.
The frontal ornament was predominantly modeled in separate subtools and then Dynameshed into the main body of the lute.
The knot-like shape, for example, is an actual spline knot from Maya that was converted into geometry.
I doubt that I could have achieved a clean enough result through masking or sculpting.
In the detailing stage, it is very important to give it a handcrafted look on high poly. I think micro-details play a major role in what may be called “the narrative” of the props.
They convey a lot of information about the physical properties and history of the material.
For example, I neglected any detailing of the holes on the leather part, giving it a machine-made impression. This was corrected in the textures.
The more detailed the highpoly is, the more details are baked onto maps, making it easier to texture. In sculpting micro-detalization, I relied again on Morph targets and layers as tools to control the depth of microdetails.
My main conclusion regarding topology in this project is that creating a personal project is not the same as developing an in-game model, so the approach should be different. The end goal for a personal project is a picture of the highest artistic value one can achieve, therefore it makes sense to allocate resources, including geometry, as long as they continue to contribute to the end result.
The entire prop is predominantly a soft surface, which is why an evenly-spaced edge flow is crucial here. Advice from FlippedNormals was somewhat helpful in this context.
The approach is to initially retopologize details as if they were the year 1996 and then continue dividing each section in two as long as it makes sense. However, smoothing may be required afterward.
UVs and Baking
The UV is pretty straightforward, without any optimization or redistribution of texels. There is some distortion on the UVs, but I noticed no effect on texturing. Baking was done in Marmoset Toolbag, and there is not much to say about it.
As for baking, this model was relatively simple. The maps were initially baked in 4k resolution, which was later changed to 8k in the final stage of production. I recommend backing the thickness map as well since I discovered it to be a powerful tool in texturing.
It was a challenge far beyond my expectations to create textures for this asset. Only after receiving the second or third feedback from my mentor Dmitrii Grachev did I start to realize that the reference is extremely rich in color variations.
In texturing, I don’t use any default materials or smart materials, but I discovered Megascans to be a powerful tool.
Leather is the first area where I utilized Quixel Megascans to compensate for the lack of creasing on the material.
As for the other parts, it followed a standard procedure of defining color variation, roughness, and height (if applicable) values and reproducing them on the texture, predominantly using standard grunges and tools.
Aside from what I saw on the reference, I made a glue transaction between the wood and leather materials.
Among all the wooden parts, the hardest was the neck. In this area, the wood exhibits its most natural color. Initially, a bare albedo was assembled from various parts of a different wood photo, so it could at least remotely resemble the wood from the reference.
Afterward, all the color variations were added. Detailing here relied on projecting Megascans more than anywhere else. This involves surface damage, roughness, and many color variations.
One important tool I didn’t know before this project is passthrough passes across the entire model, which significantly adds cohesion to the whole stack.
In the gif above, there is a green passthrough layer that goes across the entire model and, in a way, unifies its colors.
There are other similar layers that blend in cold or warm colors, change saturation, and so on.
It makes sense to blend several variations of the same effect for it to be more convincing, and this applies to dust as well.
In the gif above, it is shown how dust is being assembled from four layers, each different in color and grunge. The last layer dims the overall effect and makes the colors colder.
The final overall approach to renders was to recreate photos from references but with no restraints imposed by physics. The overall shape of the props doesn’t really fit to be put on a shadow catcher, and a stand wouldn’t allow me to render the back view.
There is an opinion that purely white lights shouldn’t be used in rendering, but I am not certain about it, so I use them.
As for chromatic lights, I keep them subtle, as I don’t like the idea that the light can significantly change the texture color.
The tone mapping is ACES.
I would like to thank Dmitrtii Grachev for extensive tutoring and feedback on this project.