HDRP camera stacking

[graceful helm]

Hello.
I am working on a mech game like Titanfall.

The Player or pilot sits inside a "closed" cockpit. The environment is visible though monitors.

The problem I have:

• render the cokpit on top of everything (above the mech arms and weapons; see images)
• make the players/pilots FOV independent from the environment view.

Now I have NO idea how to make it work in HDRP. Camera stacking is not supported OR it is way too expensive.

I already tried a multiple camera set up but it ruins the performance - I don't want to use it.
I tried custom passes to render the cockpit on top of the world (without success) but the problem of the pilot FOV would still exists.
I tried a fake cockpit with a rendered cockpit image set in screenspace-overlay but it looks not really realistic, the other UI elements can't placed in worldspace anymore and it won't affected by environment lighting...

Any ideas are highly appreciated - I am close to losing my mind...

Thanks

(The images shows a prototype scene and not exactly the actual models)

[weary dune]

What was the issue with custom passes?

You can fake firstperson fov pretty effectively just by scaling meshes relative to camera's forward axis

[graceful helm]

I am too stupid to get it working.
Whatever I do it wont render ontop of everything

I'd like to avoid scaling meshes

[weary dune]

Did you compare to examples?
Such as these
Why avoid scaling?

[graceful helm]

lighting issues

[weary dune]

More specifically?

[graceful helm]

close light sources have no or wrong impact on the model

[weary dune]

This would also be an issue with another camera with a different fov, because the position would no longer match

Forced render order makes the problem worse because there's nothing to prevent the viewmodels from clipping into the lights (or into other geometry messing up shadowcasting also)

The usual solution is to make the lighting vague enough that these mismatches will not be noticeable
https://youtu.be/11sLIyw0pWQ?t=1436
(This is made for unreal but the concepts are the same)

[graceful helm]

Probably yes. But I'd like to avoid a two camera setup anyway

[weary dune]

Yes, but I don't think there is an option to get "fov" without either scaling the meshes in some way or using another camera
Which also basically does fov by scaling the meshes it renders

[graceful helm]

I just read something about a fullscreen shader for the fov part in the custom pass.

Then I could separate the 'fov part' from 'render models above everything'.

[graceful helm]

And those linked videos are amazing.
Definitely worth a watch and I'll try it out tomorrow.

Thanks for this

[weary dune]

That sounds convenient but from what I understand the lighting mismatch and other issues rising from trying to mix the separate fov models will be there no matter how the fov is achieved
But I could be wrong

I think older games had less issues with it because lighting and shadows have gotten significantly more complex and "realistic" each generation, so the techniques that used to work for viewmodels have become somewhat incompatible

[graceful helm]

'we' will see. I have a couple of ideas now that I will try out tomorrow - thanks to you:)

Really appreciate your time and help

[weary dune]

Let me know how it goes

Examining references could be helpful too
I'd wager Titanfall games for example probably fake light interaction and shadows with the mech viewmodel parts, or do them in some very vague way

[graceful helm]

Surprisingly I found a 3rd person mod so I could check how the developers make things work.

The titan gameplay seems to be an actual full body controller - no fps arms and stuff.

I already tried fake light interactions with my rendered Cockpit approach.
I failed miserably....

[weary dune]

The difference between "actual full body" and "fps arms" is more or less arbitrary

[graceful helm]

Hey,
a small update in case you're interested:
I managed to make it work. (The image is just a prototype and not the final result I'd like to archive)
I tried many things and ended up using a two camera setup (sadly).

It's working.
However I'm sure that I'll switch to URP soon to use the native camera layer setup.
It's definitely better to have two cameras in URP rather than in HDRP....

Thanks again for your time and help

[weary dune]

Performance wise camera stacking is at least feasible in URP, although still more expensive than just rendering objects
Bigger issue usually is that multi-camera setups are totally isolated from each other in shadowcasting, and shadowless lights don't look great either if they come too close

[graceful helm]

Yep. And because the "Cockpit view" is a big part in my project I better switch to URP

The Cockpit interior is excluded from the environment lighting (lore wise the environment is visible through screens). So shadow casting is no issue in my case

[weary dune]

Why isn't rendering it via custom pass not an option again?

Seems like there'd be no downside

[graceful helm]

The Custom pass for a one camera setup didn't work on my side. The Cockpit mesh wasn't drawn on top of everything else.

And I'm actually using a custom pass to render the screens 😉

It's like the Holdout shadernode in the Blender

[weary dune]

There should be a way to get it working, so it's not like you're locked out of the option
In URP your first option would be to use Render Objects to render it in front to the same result it should be with Custom Pass, rather than a whole another camera

A camera stack can be convenient if there's a lot of different things to render that would be tedious to do with Render Objects, or if camera features like frustum culling or excluding post processing are needed
But otherwise it can be a bit wasteful or less flexible