I know it only works in screen space.
Therefore it should only update the parts of the probe that are visible from that angle ( and not the whole probe, otherwise it would be useless ).
So that when you rotate your view you still have that info from that previous camera/view angle in the probes. The more you move and rotate your view around, the more info you gather into the probes.
Without going into how useful it would be in practice - I think there would issues with mapping the on screen result to probes without extra occlusion data. DDGI uses triangle raytracing for fixing issues with irradiance grid. Eevee has no built in functionality like that. Personally Iâd like background updating grid, but this doesnât seem to be possible to keep it on background without affecting viewport. Thereâs also a pretty big cost to baking in Eevee compared to just the extra rendering cost of 1 cubemap for example, that would need to be addressed somehow for any dynamic probe solution.
I think that the other guys version is just a basic barebones version of 0451âs
Uses the glossy method via shaders, no code probably, so cant be ray-tracingâŚ
@Sleeper
Yeah, itâs the same thing, although Clement redid everything from scratch (also for the sake of clarity sake, my patch is mostly reusing modified duplication of his Eevee code also for the base functionality).
Wrote what I could gather about it from browsing the Eevee next branch code thatâs available a while back, and how it differs from what I did:
I took a quick look at whatâs available of the Eevee next source and the pure screen space raytracing part seems pretty similar. Reusing reflection rays (with some extra tweaks). Bigger change is denoising and the bilateral filter at the end of denoising uses variance estimate from the ray resolve step. Canât comment on how it fits with the other parts of the rendering or how up to date the code that is available is.
This is just from whatâs visible currently and with no way to test it out.
@hrijul_dey
I havenât tried it out, but it seems similar to what I did at first with the addon, so SSR âremappedâ to do diffuse lighting instead of specular reflections. Mine had some attempted approximations to make it behave more diffuse like, but they could cause artifacts and had a pretty sizeable performance impact. Might be worth to check it out if you need something that doesnât come in a modified build and the old addon isnât suitable for.
theyâre doing a full rewrite of eevee so probably everything is written different and they were doing a lot of optimizing in the shader size of things so they would be lighter which leads me to believe the full code will need to be re done for ssgi ⌠problem is ssgi via main branch is supposed to be coming in 3.3 perhaps so 0451 is kinda at an impasse
Iâm aware thereâs an Eevee re-write going on behind the scenes and itâs not terribly surprising that would mess up custom builds that modify Eeveeâs behaviour, but I thought 0451 meant there were issues with the quality or something that made them prefer Cyclesâ output over Eeveeâs for newer Blender versions.
Where did you hear SSGI is supposed to be coming in 3.3? Iâd love to see it added to the Blender main, but thatâs so soon - theyâre bcon3, bug fixing only and I havenât seen anything about about SSGI making it for that release. I hope youâre right though.
But yeah, if SSGI does make it into the main branch in 3.3, there wonât be much need for 0451 to continue with their version, and not much need to push it into 3.2.
There performance issues on 3.2 (at least there were last time I had time to try it out). Documented performance regression is mainly shader compilation with bump nodes. Most of my personal files use a nodegroup with bump nodes as a replacement for normal node, due to the painfully slow normal tangent calculation otherwise.
In addition changes to how the shader to RGB node is handled now in 3.2 broke most of my NPR materials.
I was using 3.2 with only Cycles for a bit, but finally decided to go back to 3.1.2 for daily usage. For my use case under 3.2, a lot of things are either now broken or almost impossible to work on due to performance issues.
Overall the situation on updating for 3.2 otherwise is pretty much as described. The recent rewrites in 3.2 also functionally wiped out a lot of my notes that I did at the beginning of the year for all the planned changes with my SSGI version (mainly kept as comments in a separate patch). The reality comes down to that Iâm unable to keep up with the main branch development unless I get enough free time continuously work on bigger changes. Iâve found that itâs incredibly inefficient to try to implement large changes if I have to do them with longer pauses in between, since I lose track of all the components that affect each other.
For the official implementation in Eevee Next I think the goal was for 3.5, but I havenât kept up closely with the development updates recently.
thats my apologies i forgot that 3.3 was about to come out i thought that was the next one my apologies the updates are showing a lot of stuff done and there was a decent update to the dev blog this week that have many of the âneed to havesâ getting there and a lot done
That actually explains a lot. I switched computers right when 3.2 came out, so it wasnât clear if that was a blender thing, a new computer thing, or just something I was imagining. I havenât done any âseriousâ rendering since 3.2 came out (just test renders of some modeling/sculpting stuff) and if I did, Iâd probably be opening it with the SSGI version anyway, but there were definitely a couple scenes that seemed to be running hotter than they shouldâve for for the materials/meshes in question.
Thatâs too bad about your NPR materials though. I should probably check some of my shaders.
Does this actually work in material preview mode?
( As itâs using screen space ray tracing just like SSR which also works for material preview. )
And does it work in both Eevee and Cycles material preview?
And is there a way to turn it of in material preview ( enable for rendered, disabled for material preview ), as sometimes you might want/need a more shade less look to see how the material âworksâ.