- New sub-surface scattering algorithm: A new, more accurate way of calculating SSS has been added. Unlike the current empirical BSSRDF method based on diffusion theory, this new method actually traces below the surface with a real random walk and makes no assumptions about the geometry being locally flat. This means it can take into account anisotropic scattering like brute-force volume rendering and produces much better results around concavities and small details. It can also be substantially faster for large scattering radius (i.e. large mean free path) compared to the old method. On the other hand, the new method can be slower in dense media (i.e. small mfp), does not support
sss_setnamefor blending two surfaces together, may require redialing materials to achieve a similar look, and is more sensitive to non-closed meshes, "mouth bags", and internal geometry potentially casting shadows. This new algorithm is exposed in the
standard_surfaceshader via the new parameters
subsurface_type(with enum values
gfrom -1.0 to +1.0). The default is to use the old empirical diffusion method in order not to break the look of existing scenes.
- Car paint shader: We are now shipping a dedicated shader for car paint, which can be thought of as the combination of a simplified version of the
flakesshaders. This shader can create a wide range of car paint looks without having to connect several nodes. For example, a pearlescent effect can be easily added to both the specular and flakes layers by simply tweaking a few parameters such as
flake_flip_flop. An arbitrary number of layers of flakes can be used (
flake_layers). The flakes at a deep layer are covered by the ones closer to the surface and more tinted by pigments (specified by the
- Subdivision frustum culling: Subdivision patches outside the view or dicing camera frustum will not be subdivided. This is useful for any extended surface that is only partially visible as only the directly visible part will be subdivided. Similarly, no subdivision work will happen if a mesh is not directly visible. This can be turned on globally by setting
options.subdiv_frustum_culling trueand can be turned off for specific meshes with
polymesh.subdiv_frustum_ignore true. The global
options.subdiv_frustum_paddingadds a world space padding to the frustum that can be increased as needed to minimize artifacts from out-of-view objects in cast shadows, reflections, etc. Note that motion blur is not yet taken into account and moving objects might require some additional padding.
- Improved accuracy of UV coords: Mesh UV coordinates are now handled with higher numerical precision, fixing jagged artifacts that sometimes appeared when using high-resolution UDIM textures over very wide UV ranges.
- Improved volume sampling of low-spread lights: Low-spread quad and disk lights should now produce less noise and show increased performance when participating in atmospheric scattering effects.
- Improved procedural namespace memory usage: Reduced memory used by procedural namespacing by about 28KB per procedural primitive, so that procedurals now use very little additional memory. In a scene with 100K procedurals, this gave about a 2.7GB reduction in memory use.
- Faster opacity masks: Texture map-based opacity masks will now be faster to render. Testing indicates around 3-13% faster.
- Faster IPR: The message logging system has been optimized, resulting in about 10% faster performance in IPR, like while moving the camera. In addition, the IPR mode in
kickhas been made substantially more responsive.
AiNodeDestroy: we have optimized removal time for nodes contained in procedurals, greatly reducing the shutdown time at the end of a render in complex scenes.
triplanarshader: Texture filtering in the
triplanarshader has been improved, giving better antialiasing and up to a 2x speedup, specially when using high-resolution texture maps.
- Celullar option in
triplanarshader now supports projection through Voronoi cells using the new
cellparameter. The rotation angle of the projected texture for each cell can be controlled with the
cell_rotateparameter. Cells can be smoothly blended using the
sizeparameter is replaced by the
densityparameter, which makes it easy to control the size and number of flakes. Alpha channel can be used as a mask. The new
flakesshader supports non-disc shapes and 3D flakes, which are useful to render gemstone inclusions like goldstone, for example.
shadow_matteshader: We have revamped and simplified the shader to make it easier to use, and fixed a number of long-standing issues: Indirect illumination now fills the global
specular_indirectAOVs, so we have removed the shader’s (confusingly named)
indirect_specularAOVs. Self-reflections are no longer rendered. A new
specular_IORparameter was added that controls Fresnel reflection. Parameters
background_typewere removed. The new enum parameter
backgroundcan be set to either
background_color, which allows to connect a specific texture in the
background_colorparameter slot. Parameter
alpha_maskwas added to control whether the alpha must be opaque or if it has to contain the shadow mask.
- Support for more OSL attributes: OSL shaders now support
getattribute()lookups of standard camera attributes (e.g.
camera:resolution, etc) as well as the geometry attributes
- Transmit AOVs and Alpha: The
standard_surfaceshader with transmission can now pass through AOVs, by enabling the
transmit_aovsparameter. If the background is transparent, then the transmissive surface will become transparent so that it can be composited over another background. Light path expression AOVs will be passed through, so that for example a diffuse surface seen through a transmissive surface will end up in the
diffuseAOV. Other AOVs can also be passed straight through (without any opacity blending), which can be used for creating masks for example.
- Improved multi-threaded render time stats: Render times when using more than one thread did not really work. We have improved this so that render times are now much more reliable and useful and can now be confidently used to determine what parts of Arnold are the most expensive. In particular, the subdivision and displacement times will now show how much total time was used as well as what fraction of that time was spent with threads unable to do useful work. This "threads blocked" time can often be lowered by using larger buckets or the
- Custom procedural namespaces: Procedurals can now declare a custom namespace using the new
namespaceparameter. This custom namespace can be used instead of the procedural name, to reference contents through absolute or relative paths. Multiple procedurals can share the same namespace by using the same custom name. Also, they can declare an empty name and they will use the global namespace. (#6085)
kick: When using
kick -turn, the
-turn_smoothflag can be added to smoothly start and stop the movement as the original position is reached with a cubic ramp.
kick -laovs file.asswill display a list of all the built-in AOVs and all the AOVs registered by this .ass file.
- Added cputime heatmap view to
kick: When using
kickyou can now toggle between viewing kicks default output and a cputime heatmap with the
Tkey. The mapping of the heat map can be scaled with the
- Support for wasd keys in
kick -ipr m: Running
kickin Maya-style IPR mode with
-ipr mnow supports "wasd" style keyboard movement. This was previously only available in Quake-style mode,
maketxversion info: The custom
maketxbinary that ships with Arnold now reports, both in the command-line and in the embedded EXR headers, that it was built specifically for "OpenImageIO-Arnold", to distinguish it from the official "OpenImageIO" one.
maketxcolor spaces: The custom
maketxthat ships with Arnold now supports OCIO and SynColor (when available) through the
AiMakeTxreports info messages:
AiMakeTxnow also prints out OIIO informational messages (generated in verbose mode, for instance) in addition to errors.
AiMakeTxreleases input file lock sooner:
AiMakeTxwill now close the input texture as soon as possible instead of waiting for all the maketx jobs to finish.
maketxoptimized flags: Both
maketxnow always run with the flags
--monochrome-detect --opaque-detect --constant-color-detect --fixnan box3 --oiio, which can result in smaller .tx files that are faster to load and take less memory.
- Report when textures are changed during render: The log files now report when texture modifications during a render cause a texture read error, which can happen in certain pipelines.
- OCIO color space family support: The OCIO color manager now implements color space enumeration by family. This is useful for UI drop down organization.
- OCIO view/display enumeration: The OCIO color manager can now enumerate view/display combinations using the "View (Display)" family. This lets client programs filter color spaces when only a display transform is appropriate.
- .ass metadata from compressed files: We now support loading metadata from
.ass.gzfiles through the
- Upgraded to OIIO 1.7.17: OpenImageIO has been upgraded to OIIO 1.7.17.
- Color space family enumeration: Existing color space families for the current config can be enumerated using the new API methods
AiColorManagerGetFamilyNameByIndex. The addition of these new API methods requires any existing custom color managers (which we know are very rare) to be recompiled.
AiAOVSampleIteratorGetPixel(): Custom filters can now determine what pixel is being filtered with the new API method
transparentLPE label: When setting the
transparentLPE label on a BSDF, the surface will act as if it is transparent and pass through AOVs. This would typically be used for transmission BSDFs, as it is in the
- Deprecated API warnings: Defining
AI_ENABLE_DEPRECATION_WARNINGSwill cause the compiler to emit warnings if deprecated Arnold API functions are used.
- Random walk SSS closure: The new random walk SSS algorithm is exposed in the C++ API as
AiClosureRandomWalkBSSRDF(). The corresponding OSL closure is
autotiledisabled by default: We found that the
autotilecode in OIIO does not scale with high-resolution textures. In order to avoid very slow loading of untiled textures (such as JPEG) when autotile was enabled, we have now changed the
autotiledefault setting to 0, which effectively disables it. In very rare cases, when rendering with a large number of high-resolution untiled textures, this change might degrade performance as the texture cache will blow up. The real solution is to never use untiled files, and instead convert all untiled textures to .tx tiled files.
maketxdependencies: The custom
maketxthat ships with Arnold now depends dynamically on
libai.soand optionally on
syncolor_shader.so, therefore to work correctly it needs to be run from its installation folder, like
kick, or alternatively the new dependencies should be copied to where
maketxis running from.
- Light groups and volume shading: Just like with surface shapes, Arnold now obeys light group assignment on volume shapes and surfaces with volumetric interiors. While in many cases desirable, this can produce an unexpected change in the final image in scenes with light group assignments.
flakesshader: It was not easy to control the number of flakes with the
scaleparameters because they were mutually dependent. Now this can be easily done using the new
densityparameter. The shape of each flake has been changed from disc to Voronoi cell, which is more suitable to render inclusions of gem stones. The shader output type has been changed from RGB to RGBA to support a mask.
motionvectorAOV: The motion vector scaling factor in the built-in
motionvectorAOV has changed. This was required to fix a bug that caused zero motion vectors for certain shutter positions. The output from the
motion_vectorshader is unchanged: it can be used as a workaround in your old scenes if you require the previous scaling.
shadow_mattechanges: The shader AOVs
indirect_specularwere removed, since the shader now fills the corresponding built-in AOVs. Parameters
background_typewere removed. Specular reflection is now affected by Fresnel. To roll back to the previous specular behaviour, set
specular_IORto a high value like 100, which effectively disables the Fresnel effect. See notes in the enhancements section above.
- #3319 Alpha not fully opaque in output images