Oni2:Big/Earthquake

From OniGalore
Jump to navigation Jump to search
Earthquake (Samurai Shodown) ZBrush model (clay).jpg

Earthquake is a character from the Samurai Shodown fighter games.

Probably the biggest human character in the history of ever.

So naturally he's a huge (^_^) reference for Oni2's Mr. Big.

There's a really cool ZBrush model of him by Arnaud Kotelnikoff [1]

(released in November 2011, T-Pose version in June 2021)

And that's about all you need to know for now.

P.S. Mr. Big was imagined before I became aware of Earthquake. geyser

Earthquake (Samurai Shodown) ZBrush model (T-pose).jpg

Adapting the ZBrush model

Arnaud Kotelnikoff's Earthquake is not the only appealing XXL character model out there* but he stands out in many respects as an almost game-ready asset: sculpted muscles and veins readily convertible to a normal map, a distinctive badass face, fully modeled eyes and mouth... He also represents an original enough take on the copyrighted character, so that he will not be instantly recognizable as "Earthquake" (except to those familiar with Kotelnikoff's model).

The original model is available as a free sample to anyone using or trying out ZBrush, and the T-pose variant was made available separately rather than bundled with a ZBrush product. However, as mentioned on ZBrush Central, "the Earthquake character is under the Samurai Shodown copyright", meaning that the model cannot be freely used as-is without a special agreement with SNK. Basically he needs to drop the tattoos and get some new clothes.

The T-posed variant makes rigging and animating a lot easier than for the original pose. However, a quality result is not trivial to obtain, and calls for a tutorial on the following tasks: slight topology and T-pose adjustments, weight transfer between LODs, normal map baking, and possibly UV unwrapping and diffuse texture baking as well.


*Searching Sketchfab and other resources for "sumo" and such turns up, e.g., THIS remake of Street Fighter's E. Honda. (Also there is always the option of getting really good at 3D modeling and Making Your Own. -_-)

Programs needed

As of now [02:45, 30 May 2022 (CEST)] there is no third-party tool capable of reading ZBrush's proprietary ZTL (ZBrush Tool) format, therefore you need ZBrush to access the mesh data and convert it to an exchange format such as OBJ or FBX. ZBrush is an expensive app ($39/mo, $359/y, $895 perpetual), and the lighter/cheaper ZBrushCore ($10/mo) and ZBrushCore Mini (free) cannot open ZTL files, leaving trial as the only sensible option for a non-professional (a trial subscription to "Maxon One" lasts 14 days).

Obviously ZBrush is a very capable tool, but it also has a unique interface (not at all intuitive unless you are already familiar with 3D sculpting). The learning curve is such that you aren't too likely to do any advanced modeling within the 14 days allotted for the free trial, and you are better off just grabbing all the meshes and exporting them to OBJ or FBX so that you can continue the work in other 3D apps, such as Blender.

Blender is practically the only other app that you will need, apart from maybe Mixamo.com (a sensible alternative to Blender when it comes to auto-rigging).

Model overview

SDiv 1 2 3 4 5 6
Earthquake_82_optim
verts 1,209 4,852 19,402 77,602 310,402 1,241,602
polys 1,218 4,850 19,400 77,600 310,400 1,241,600
tris 2,414 9,700 38,800 155,200 620,800 2,483,200
Eyes (sculpted UV spheres)
verts 228 964 3,844 15,364 61,444 N/A
polys 256 960 3,840 15,360 61,440 N/A
tris 448 1,920 7,680 30,720 122,880 N/A
Teeth (32 teeth + gums)
verts 312 1,044 3,972 15,684 62,532 N/A
polys 244 976 3,904 15,616 62,464 N/A
tris 488 1,952 7,808 31,232 124,928 N/A
Kimono (shirt + pants)
verts 3,776 15,132 60,552 242,232 968,952 N/A
polys 3,786 15,140 60,560 242,240 968,960 N/A
tris 7,568 30,280 121,120 484,480 1,937,920 N/A
Lace (shirt decoration)
verts 144 564 2,244 N/A N/A N/A
polys 140 560 2,240 N/A N/A N/A
tris 280 1,120 4,480 N/A N/A N/A
Taping (around right hand)
verts 144 592 2,380 9,532 38,140 152,572
polys 150 596 2,384 9,536 38,144 152,576
tris 296 1,192 4,768 19,072 76,288 305,152
Belt (pants decoration)
verts 136 538 2,146 8,578 34,306 137,218
polys 134 536 2,144 8,576 34,304 137,216
tris 268 1,072 4,288 17,152 68,608 274,432
Shoes
verts 514 2,056 8,212 32,836 131,332 525,316
polys 516 2,052 8,208 32,832 131,328 525,312
tris 1,020 4,104 16,416 65,664 262,656 1,050,624

The original model consists of eight meshes (or "brushes"), the polygon and vertex count is shown on the right. There are no textures (skin color variations, tattoos etc are per-vertex color attributes).

The T-posed model mostly has the same topology as the original, but with the following differences:

  • The shirt becomes a separate mesh, "Kimono1" (and the pants stay as "Kimono").
  • The pants have an additional subdivision level (low-poly: 598 verts, 608 polys, 1208 tris).
  • The "Taping" mesh covers both hands (the vertex and polygon counts of all LODs are doubled).
  • The eyes have no subdivision levels (only the high-detail mesh is present).

With all due respect to ZBrush and to the author, the T-posed model has several issues as compared to the original, some of which pose a problem for rigging:

  1. The T-posed clothes have negative thickness in some areas, so the backface of the inner surface pokes through the outer surface and vice-versa. Easily fixed (and also irrelevant since we are discarding the clothes).
  2. The backface of the T-posed mouth cavity slightly pokes through the upper lip as well (for the higher LODs). Does not cause problems for normal map generation.
  3. The pose is not exactly symmetric when it comes to the hands (see, e.g., the index fingers in the above screenshot). Thankfully the body has symmetric topology, and the sculpting (e.g., veins) is symmetric as well, so the symmetry can easily be restored.
  4. The central edge loop is misplaced (rotated slightly away from the YZ plane). The adjacent polygons end up overlapping and facing inwards at the highest LODs, so the normal maps cannot be generated properly. Luckily it is only this one loop that is messed up, and it is easily recentered.
  5. The arms are angled much lower than for a typical T-pose and, given their thickness, there is a big fold in the armpit region: the topology is inaccurate for the higher LODs, hence (again) bad normap mapping.

There are also a few general issues with the body mesh, which occur not just for the T-posed version but for the original model as well.

  1. The geometry is slightly distorted at mesh poles (tiny quads): fingernails, shoulderplates, mouth corners and especially ear lobes.
  2. The fold between the pinky and ring finger is messy (apparently a low-poly sculpt defect that got worse after subdivision). For the original pose, the left fist has messy folds between all the fingers.
  3. The creases below the buttocks have slightly messy topology, but not as much as for the armpits or the crease between the ring finger and pinky.
  4. Generally the topology of lower LODs does not conform to the body shape, with lots of non-planar quads. Probably the low-poly topology should be re-done from scratch.
  5. The UV unwrap is also non-symmetric (probably because it was done for the original pose), and far from optimal in the head area (too small for the amount of sculpting detail). Possibly a new unwrap would be better.


Adjusting the T-pose

The Earthquake is centered on the center-of-gravity. Raise the mesh by 1.06 meters and bake the offset into the mesh.

The central edge loop of the body mesh must be fixed for every LOD. In Blender, switch to Edit mode, select an edge near the YZ plane, then pick Select/Select_Loops/Edge_Loops, and scale the loop by a factor 0 in the X region, with the 3D cursor (world center) as pivot point. The central edge is harder to identify for the highest LOD.

Fix the left hand by picking Mesh/Mirror/Restore_Symmetry and choosing "Negative side" as Source. This leaves only the crease between the ring fingers and pinkies, the armpit folds, and the backface of the mouth cavity.

The armpits are the most critical and least trivial fix. We are going to auto-rig the mesh with Mixamo, raise the arms of the auto-rigged mesh until the armpits are exposed, so that the high-LOD topology can be fixed and flawless normap maps can be baked on the whole mesh.

Mixamo's auto-rigger cannot directly handle the detailed sculpt body meshes: if you attempt to upload the higher LOD, the process will fail either immediately upon upload or while rigging.

In the problematic armpit regions, Mixamo does not attribute weights correctly (chest sides are affected by arm movement), but this can be remedied with weight painting.

Remeshing