How to Cull Bounds of Instances in GeoNodes

I’m using this pre made camera culling geo nodes setup by caramel https://bbbn19.gumroad.com/l/vdrdfy?layout=profile
But it culls based off of origin point so it isn’t accurate when culling larger objects. How would one implement the Instance Bounds node to cull based on the bounds of instances instead?

cull blend

hey, check it out: https://blueish0930.gumroad.com/l/vsxvj
It can prune instances from camera based on its AABB, OBB or Convex Hull. And can also produce non-overlapping instances layouts.

Here is article on how to do it:

Basically you need to check each 8 corner points of bbox with frustum planes

https://imgur.com/a/oTgIdjd

Blend file:

Dude, You are a LEGEND! :angry:

why bro gotta use the ai :face_with_diagonal_mouth:
omg have you seen the node setup?! 10$ and looking like that is crazy. wonder if the whole thing is just ai generated.

:sweat_smile:definitely not ai. If you ever saw the tutorial, you wouldn’t say that
Here is x link: https://x.com/blueish0930 , i don’t think ai can do the node work right now

I use ai because i’m not good at building a nice scene to demonstrate it. SO i just let GPT to generate the covers for me(i’m too lazy)

well if one thing is ai then nothing can stop me from assuming everything is. I watched some part of the tutorial but you never got into the nodes so I didn’t see the exact setup till that one screenshot. and AI tends to fumble node setups.

Imagine how messy it would be without node groups? It will be a shit and you probably for everything after a few days. It’s always good to keep the node neat. And, why you assume this is ai generated? I never see ai can generate such a complex node setup! AI can code a new system for you but i don’t think it can figure out the geometry node solution. Otherwise all gn users and houdini users will be fired​:sweat_smile:

I didn’t got into the node groups because this is a guide tutorial to help people know how to use. It’s not a tutorial to teach people how to build it. It’s a complicated setup, require you know about normalized device coordinate, collision detection and collision response algorithms like sweep and prune, sweep axis theory(SAP) and even GJK, MPR, EPA. Even if i show it ,you won’t understand the logic behind it

:point_down:

To expand on that, here’s an IRL example: If someone is sellig me a car but the picture they show me of the car I’m buying is a 10 second doodle, I expect the same amount of effort in the doodle has been put into making the car. Like there’s a reason companies spend millions on advertisements alone.

Not that hard for it to create a giant spaghetti setup tbh.

ya i got that cause then anyone could just recreate it for free. Just saying that I only assumed the rest was AI was because I saw the gumroad setup pic and didn’t expect a 10$ setup to look like that. I’m use to seeing complex yet organized setups on 0$ files.