Main change are:
- the fresnel LUT is separated from the main GGX LUT.
- LUTs use sqrt(1.0 - NV) as roughness remapping. Improving precision and
removes needs for acos().
- LTC LUT is normalized by matrix middle component. Improving precision.
We are still ditching the specular intensity of SSR (ssr_data.xyz).
But at least now there is some comment about it.
See T61704 for user reports on that matter.
Comments with the blessing of Clément Foucault.
Some of Eevee's Bloom defaults are not very good for physically based rendering. This patches addresses this issue.
This picture shows one of the problems with current default. Bloom looks very foggy:
{F6280495}
Even worse, light emitters much dimmer than the Sun can make everything equally hazy if Clamp is set to 1.0 and intensity to 0.8 (current default). Artists often forget to adjust Clamp value and do not know what value to use for realistic intensity. Also, currently both Clamp and Intensity do not have good UI ranges. This is why often Eevee renders end up very hazy and bloom often does not look right.
Bloom effect plays important role to help to distinguish between bright and relatively dim light sources. With current defaults this is broken because Clamp set to 1.0. Also, it cannot be disabled if set to 0 like expected. This patch fixes this and sets it to 0 by default. If users need to clamp, they can do so easily with UI range up to 1000. This range is good enough for most cases and provides enough precision to control lower values, and the highest value helps to limit bloom from the Sun if necessary and will leave untouched most other light emitters. If needed, much higher values for Clamp can be entered manually up to 100000. 10000 is still affects the Sun, but up to 100000 highest limit allows to clamp anything that is much brighter than the Sun if user needs to limit bloom in such cases (for example, bright explosion in the sky or anything else very bright).
I propose new default for bloom Intensity - 0.05 and UI range to suggests realistic values. Bloom Intensity > 0.1 is not realistic for clean lens but the user can enter manually much larger values if needed.
For comparison, here is a my own photo with and without bloom caused by the Sun (on second photo the Sun was occluded with an object).
{F6280500}
{F6280492}
In real life bloom is much more subtle and does not look hazy. If Clamp is disabled, then out of 0.1, 0.05 and 0.025 values I have tried, 0.05 looks most similar to the photo. Here is test render with and without bloom with the Sun in similar position like on the photo:
{F6280496}
{F6280494}
Using color probe 27x27 I compared lightness below the horizon under the Sun. In rendered by Eevee images lightness difference was 17. In case of the photos lightness difference in similar place was 11. I then compared leftmost spot (also below the horizon) and lightness difference was approximately 2 between two photos and 1 between rendered images. In other words, with these settings bloom effect is not too strong and is not too weak. Visually it may seem like decreasing bloom intensity may increase photorealism, but then bloom effect would be too localized even for the Sun.
Besides this single test, I tested in many other scenes as well, with and without the Sun, with different HDRIs, and as far as I can tell 0.05 intensity turned out to be good default - it produces bloom strong enough to be noticeable and not too hazy.
In Cycles shutter default value is 0.50, so for consistency set to 0.5 by default in Eevee too. Besides, 0.5 is typical standard for real cameras, and values higher than 0.5 usually are needed only if very strong motion blur is desired.
Here is summary of all changes:
Bloom Intensity: 0.8 > 0.05
Bloom Intensity UI range: 0-10 > 0-0.1
Bloom Clamp: 1.0 > 0.0 (disabled by default)
Bloom Clamp manual range: 0-1000 > 0-100000
Bloom Clamp UI range: 0-10 > 0-1000
Shutter: 1.0 > 0.5
This patch is related to the discussion in this thread, there are more examples of what bloom will look like with 0.05 intensity by me and others:
https://devtalk.blender.org/t/eevee-needs-to-have-physically-based-defaults/4700
Reviewers: fclem
Reviewed By: fclem
Subscribers: pablovazquez, billreynish, rboxman
Tags: #eevee
Differential Revision: https://developer.blender.org/D4212
This changes the bent normal effect to be a bit more subtle.
I also tuned down the bent normal blending factor so mesh faceted look may
appear more in occluded regions. this is to increase the fidelity of the
indirect lighting. This blending might be a parameter in the future.
Based the calculation on "Bent Normals and Cones in Screen-space"
by O. Klehm, T. Ritschel, E. Eisemann, H.-P. Seidel
This enables reducing the noise comming from very bright light sources
(like a sun) that can be found in distant HDRIs.
The lost energy may be replaced manually by a sunlight that compensate the
this loss.
This clamping only concerns Reflection Cubmaps and is done on all on all
of them.
Setting to 0.0 disables it (default).
This is a parameter that will make the interpolation between irradiance
cells of a same Irradiance Volume smoother, reducing the weight of the
light leaking correction factors.
It is usefull in some cases to avoid harsh lighting transition that can
happen when a sample point it near a surface.
This is an important change. Starting from now, all lights have a finite
influence radius (similar to the old sphere option for BI).
In order to avoid costly setup time, this distance is first computed
automatically based on a light threshold. The distance is computed
at the light origin and using the inverse square falloff. The setting
can be found inside the render settings panel > shadow tab.
This light threshold does not take the light shape into account an may not
suit every case. That's why we provide a per lamp override where you can
just set the cutt off distance (Light Properties Panel > Light >
Custom Distance).
The influence distance is also used as shadow far clip distance.
This influence distance does not concerns sun lights that still have a
far clip distance.
---
This change is important because it makes it possible to cull lights
an improve performance drastically in the future.
I made an empirical test with a 100% diffuse sphere and manually tweak the
lighting power of a sun lamp trying to fit cycles and eevee the best I can.
Then I plotted the result and found a rough fit to the equation and that
seems to work pretty well.
This makes it possible to tweak indirect lighting in the shader.
Only a subset of the outputs is supported and the ray depth has not exactly
the same meaning:
Is Camera : Supported.
Is Shadow : Supported.
Is Diffuse : Supported.
Is Glossy : Supported.
Is Singular : Not supported. Same as Is Glossy.
Is Reflection : Not supported. Same as Is Glossy.
Is Transmission : Not supported. Same as Is Glossy.
Ray Length : Not supported. Defaults to 1.0.
Ray Depth : Indicate the current bounce when baking the light cache.
Diffuse Depth : Same as Ray Depth but only when baking diffuse light.
Glossy Depth : Same as Ray Depth but only when baking specular light.
Transparent Depth : Not supported. Defaults to 0.
Transmission Depth : Not supported. Same as Glossy Depth.
Caveat: Is Glossy does not work with Screen Space Reflections but does work
with reflection planes (when used with SSR or not).
We have to render the world twice for that to work.
NOTE: There is a float imprecision near the focus plane
due to the current technique used for DOF. This makes the alpha channel
transparent on nearly in focus objects even when they should not.
This artifact should be fixed when the DOF will use scatter as gather for
low brightness areas.
Fix T57042 : Eevee does not render alpha when DOF is turned on
This implementation is a bit hacky but match cycles pretty close.
If pixel size is not enabled, it will use the geom shader to
compute distances between vertices. This will have a cost.
Implementation is a bit hacky in gpu_codegen to make the geom shader works
in an optional manner.
There was an issue caused by Antialiasing being done after DoF. Move TAA
after DOF and Motion Blur.
Also certain pixel with lower CoC would be spread all over the background
because the neighbooring pixel have higher CoC. So we need to apply some
bilateral filtering when downsampling. Currently we limit the influence of
neighbor pixels with a CoC inside the range [MaxCoC-2, MaxCoC].
Note to myself, next time, better check the fix before pushing it.
GL_ARB_texture_gather is defined if there is support for the extension
not only when the extension is enabled. Do this check ourself with
GPU_ARB_texture_gather define.
Original fix 822de6e9e1
The extension GL_ARB_texture_gather is reported to be supported and does
not trigger an error when enabled but the textureGater functions are not
defined.
Workaround is to disable the use of this extension on such systems.
This is more like a static optimisation when some parameters are set to 1.0
or 0.0. In theses case we use a more optimized version of the node.
This also revisit the transmission parameter behaviour to make it closer to
cycles.
