The Illusion of Motion: Making Magic with Textures in the Vertex Shader 

While these are potential things that affected the result, them existing does not improve the presentation. Ideally, a good presenter would take steps to mitigate these problems, and if they don't or couldn't, the critique is valid. An example is the guy having visuals, which definitely helped. The information is valuable and still appreciated, but communication is important in the sharing of information. Remaining agnostic to the problems of a good presentation, does not improve the presentation.

Agreed. I liked the presentation. He done better than I would if I had to speak any other language! ;) He presented some fascinating ideas I thought.

@@tnpcook: I tell you what, get a refund if you didn't like it, otherwise, STFU.

@@NeilRoy being critical of a detail isn't mutually exclusive with enjoying it. It would be absurd to dismiss the good presentation for that alone, but also disingenuous to ignore.

I like these kind of speakers most, because they know about it and try to pronounce as understandable as possible. That's if you're going with it. maybe amp up the speed. But lot better for other nonnative speakers than many native speakers.

Is it publicly acessible anywhere? Buying hardcover for few pages is not an ideal solution, especially when it is not available to be shipped to my country.

@@frajet ShaderX4 hasn't been made publicly free yet unfortunately

@@wbahnassi I guess still the hlsl implementation is on the CD instead of written as text in the book, so it wouldn't help. I've read on twitter you wrote: "But nowadays you can also do with StructuredBuffers for more flexibility than textures.". But StructuredBuffers is a bit complicated for use, especially in UE4, so I would like to stick to textures, but need to somehow bypass the 8192 verts limit, but why is it packed to the texture so inefficiently that Y = animation frame and X = vertex position? Why the texture is not packed more tightly to occupy more space of the texture and thus allowing more than 8192 verts? Isnt it a matter of simple lookup code with module in the vertex shader? What am I missing here? Thanks.

@@frajet That was referring to skinning. Texture size limit (8k or 16k) means you have up to 8k or 16k bones controlling the mesh, regardless of how many verts it has. I've never seen a character animated with more than a couple hundred bones in real game production.

​@@wbahnassi Yes I understand that, but one of the motivation in this video at 14:50 was "Huge amount of vertex data.". Which is the only thing I need, without using bones. In UE4 they've 'recently' implemented this old technique storing vertex offset positions in the texture docs.unrealengine.com/en-US/Engine/Animation/Tools/VertexAnimationTool/index.html while saying 8k is the limit of the verts stored in the texture, because of the 8k tex limit, but my question is simpler than you think. Why did they not pack texture more tightly by just changing lookup a bit? It stores frames on Y axis and verts positions on X axis, but there may be max. few hundreds of frames stored, while verts may be >8k, thus it doesnt make sense and that is why I am confused here. Then I've thought there may be some better way to send verts to gpu and reference them from vertex shader - and found your tweet about StructuredBuffers, but I've found it complicated for my purpose. I only want to pack say 20k final verts offset to texture, without any bones. Is that an issue? I see that it is doable, but somewhat I am confused with UE4's implementation limitting verts to 8k because occupying one axis by the animation frame. So I am confused why did they choose this approach as there is too big space for 8k anim frames while 8k verts limit is too little.

I can't personally think of a reason that would prevent you from doing so. In it's most complex form it shouldn't be any different than assigning data and telling the engine where to start referencing the texture from. I think this is where third party tools could really help to push the tech forward.

It would be possible, but not worth it. The whole point of baking the animation into a texture is a trade off between memory and processing. The animation is already computed. If you were to change the animation dynamically you would have to calculate it on the fly and that defeats the purpose of baking it. That is why he suggested it's use to facial animation and the such.

Besides how would you detect collision if the animation is baked? With regular skinning you can attach your colliders to your bones for dynamic collision detection, with baked animation your can only use a capsule collider or if you come up with a way to animate them following the mesh vertex animation do let us know

Wonder why nobody agreed with your opinion in a year+

That's what I thought, but to my amazement I understood like half of it!!! I am progressing :D

It was a 3dmax custom made script

XYZ to RGB, literally took the normalized value Of XYZ and used them to compose each pixel, repeat for all needed transforms and you're done

Why to do so when there are mesh shaders now which do pretty much the same, but do it properly?

@@pavelnikulin8240 newbie here, are you referring to a geometry shader?

@@EveryCrime It's more like a compute shader

It is a texture because it is literally a texture in memory.

Agree, I can't go past 5 minutes, which sucks.

It's hard to be very good at something nerdy and a good public speaker at the same time, I'd rather it be this way than the other way around. (i.e. a good speaker that doesn't know their subject well)

he is doing his best guys damn

@@peacku you're right, something is better than nothing, it just seems a shame to me that his brilliant ideas can't be articulated. I watched til the end but I fear many won't make it as far

@@giampaolomannucci8281 I recommend watching from 15:20 forward if nothing else. He talked about storing animation in textures.

While I'm a fledgling game designer at best and I respect your opinion, I dont think it's entirely useless. This seems like a solid optimization process that might turn out to be vastly superior if the tech is developed more. That and I can see indies benefiting from this method of development where you can rely less on engine specific optimization tools in an effective manner. I think if a third party comes out with a tool that streamlines this process we could be looking at something big.

@@Zaft9 Yeah I could agree with that.

Yeah why use the vertex shader when we all have 32 core CPUs and RTX 3090s at home, fuck mobiles/consoles and low powered hardware /s

Being close syntax wise to C I guess C sucks ass as well?