Відео

No, this is what makes this work so well. What I do is I feed in normal triangles the regular way in the pipeline. Then I make them thicker by adding some 4 more trianlges. Then I raymarch them in the fragment shader and discard if it hits nothing. This way you should usually get a hit within 1-3 marches.

Oh, also this would work a lot better with BVHs. I think you are doing raytracing with multiple different data structures? I would highly recommend adding this into your engine because it supports large animated scenes. However, I have not done this yet. But it is theoretically possible.

@@capslpop Really smart, I was actually thinking of a similar idea myself the other day, but couldn't figure it out fully. So you rasterize a "thicker" version of the mesh (I'm guessing by moving the vertices along their normals), then raytrace starting from the rasterized position? And you raytrace just by doing a DDA and using a triangle SDF for every triangle in the mesh? If it's that simple I will totally try implementing this into my engine, has a lot of potential!

@@frozein Yes, its that easy! I would like to see someone try it for BVHs. I think it could really revolutionize the way tiny voxel based graphics is dene. I also belive I have a few tricks up my sleeve to fix the part when triangles intersect each other.

@@capslpop Super cool. I'll try implementing it myself once final exams are done, I can do it with BVH and see the performance. You can add me on disc if you want to talk more, same username as my UA-cam.

It is rasterized then marched.

I think its like 2000 or something.

Depending on how you define it. But yea you could say that. I also just posted the code on shadertoy so you can see how it works now

It looks much better oddly enough 😄

If there are more than two particles per cell. The particles are grouped together. Also, I do a trilinear texture fetch for the G2P step to speed things up.

its called teardown :)

@@MountainLabsYT i know that game and played it but thats not the same

@@Mertiven do you mean the per-voxel lighting?

No, this is Shadertoy kinda similar though. If it was it would be like 10 times more impressive . 😅

No, instead I make some assumptions based on the geometry of voxels. First, each ray collects a cube and stores them. Then, I draw the cubes back using a higher resolution.

I was actually plaing around with that. I got it to look like oil with an iridescence surface. Looked kinda cool. ua-cam.com/video/mGTdCkr2D1E/v-deo.html

Oh and here is clear. It kinda looks like glass: ua-cam.com/video/ny_hWM-cmkg/v-deo.html

@@capslpop Oh awesome, thanks for uploading those!

For the full completed engine. I plan to have it adjustable. at medium, its probably going to be around 1-2 gb and for high level like 4-5 gb

@@capslpop 2.8 trillion voxels in 4 gb? that's some serious compression, how are you going to achieve that?

@@Meetric1 Here its litteraly jut instancing haha to the data is copied over and over again but I will be using LODs with "circular pointer array" thing that will hopefully be able to store way more data than this demo because it wont be instanced. Almost got that done and I will post it on YT. Its actually *fairly* easy to render like trilons of voxels whats hard is editing those voxel each frame with lots of updates and still keeping it fairly compressed.

interesting, thanks for the info!

Raytraced but I want to do a mixed rendering method. Actually about to implement that today.

Ayo, the gmod physics guy is here :D

the gwater guy

Mainly shadertoy is what I am using. However, I am not really using many resources. A lot of these algorithms are ones that I have come up with. For the scene with the tree and grass. I litterally itterate over every voxel and do different test functions to find out how the scene should be constructed. The tree is a fractal. But this is very slow and kinda cursed. Magica voxel has a way better editor. However, I am planning on doing a devlog explaining how I am doing all of this stuff in the last few months.

because who wouldn't want to see a lot of voxels. jk

Actually not really much. The same stuff I did back in my first devlog with using rasterization. But this time they are in chunks.

I am using a laptop 3070. However I hope to make it performant enough to run on an igpu

No, but I will add that in the future. Should be supper easy to add ray acceleration.

I still have many optimizations to make. Using this technique should be a lot faster than traditional raytracing. Much of the noise will disappear with the optimizations too. Hopefully, haha!