Stencilmap-Autoknife
(not waiting around)
---------
This technique articulates large planar "brushwork" geometry by subdividing at specified coordinates via an automated method using "stencilmaps" with Blender's knife tool, among others, which is both a more efficient and more elegant, far less vulgar approach to displacements for presentable real-time assets.
Unlike conventional heightmap displacement methods that attempt to optimise an evenly tessellated surface, this technique will focus subdivisions at scripted coordinates derived from features of the normalmap used in the material of the surface. The normalmap is automatically processed into "stencilmaps" that outline the angle differences of the normalmaps by their tangent offsets, a change in hue from the flat lavender default. Each outline is able to represent a "pool" and this is guaranteed to follow the feature of the normalmaps by both its inner and outer periphery, using ImageMagick to discern these in a sequence. Each pool is then subtracted and the normalmap is processed again without it, until all pools of features are subtracted and the image is identified as a single non-standard normalmap value and is empty. These individual pools are used each as a stencilmap, which is a 1-bit image that are unique contiguous islands of form and feature. These stencilmaps are then each turned into a "stencil-atlas" which represents each tile of the UVs as applied by the mapper as a texture, including its repetitions, scaling and angle. These are processed with Inkscape command-line and its "trace bitmap" path process, to walk vectors around the island of pixels in both an interpolated way and with an optimised result, and this is saved as an SVG file of vertices, and transformed via data management into a script of coordinates in Python for use in Blender and its knife tool. The process will assign each compatible quad its own camera which is assigned by that mesh's surface normal and the entire procedure cycling through them and triggering the stencilmap operation for whichever material header is associated with that model. Each quad is tessellated via a basic first pass which ensures the geometry is bound to the same tiling and proportion of the texture being used, which also helps to keep the result more consistent. It can not duplicate the operation of these tiles and each one undergoes its own entire process, however it will ensure that each one is identical, even if it's not aligned with the UV offset of how the texture was assigned in Radiant. This offset is handled by subtracting those coordinates from the way the stencilmaps are converted into the automation script, and in this way there remains far fewer restrictions and concerns for how mappers texture their brushwork and wish to use this technique.
The script will essentially process cuts in the target geometry by first preparing the viewport via the camera assigned to each face which is its target, setting a universal scale where each quantity of texelation can be precisely utilised via actual pixel coordinate, a universal granularity. Each stencilmap coordinate is sliced from each other neighbouring stencilmap "match" and from each direction, until the entire "pool" has been isolated. Once this is done, the first pool is cleaned and hidden, and the remaining geometry is remeshed and ready for the next stencilmap operation. Each pool degrades in quality as this process is repeated, however the largest pools will always be done first which should typically avoid messy results on prominent features. The rasterisation nature of the automation will preserve horizontal definitions and the vertical definitions will be clean on the first iteration with the largest pools, but clean vertical definitions will suffer throughout the procedure with each pass as it inevitably has less to work with. Once each stencilmap knife pass is completed, all components are recombined, optimised, remeshed using a quadratic fill which is capped at the periphery of each tile, cleaned and displaced by the material's heightmap. The result can then be imported back into Radiant, thus becoming clean presentable content that can be confidently used in much higher LOD stages. This entire process is possible to automate fully, if the mapper exports specific geometry that meets a few requirements regarding the planar and expansive explicit coordinates required by the code. Other concerns involve the nature of the normalmaps, but the process will still work even if the result may be redundant or undesirable, which would be a waste of time on the back-end but easily enough disregarded.
Quite a themepark full of rides but it's all able to be done via commandline automatically. Provided the preparations are taken and issues are negated, as mentioned below, and although it is supposed to be an iterative process, there's no reason a full map of hundreds of surfaces can't be processed at one time. It's not perfect and will never beat manual articulation by a talented mapper, but it does result in less triangles being used, combined with much cleaner results with sharp definitions instead of sludgy lumps, and an automated way to increase the quality of huge areas of maps without much of any effort. The technique is also very scalable in of itself by coupling the stencilmap pool peripheries and adjusting some other simple parameters, and I foresee much of this technique being improved in the near future in more complex ways. Ideally one day being able to handle curved surfaces and finer geometry in a seamless way. By articulating large flat surfaces by themselves and using their own textures to do it in an efficient enough way, enough to deploy in real-time and already further enhance the scalability of an entire project.
...
(issues can be fixed by editing stencilmaps, or editing geometry before or after displace.)
issues:-
- Flat normalmaps in reflective materials with small isolated details.
..."highly reflective surfaces of which the majority is flat."
- Tiling normalmaps with seamless angled tangents.
..."results in unwanted steps and fences on supposedly flat surfaces."
- Normalmaps with subtle gradients.
..."unable to resolve the stencilmap enough to create depth curvatures."
- Corners and seams mismatch.
..."these will all need to be manually scaled and welded, regardless." (for now...)
- Decal offset.
..."heavy displacements become less compatible with flat decals."
preparation:-
- Brush faces must be no less than 64x64 units, contiguous and planar.
..."they can still be on any angle."
- Maximum faces can not exceed 8192 units.
..."must be split into components."
.
.
.
;){kind=link}