Відео

ua-cam.com/video/c1jupIDfQ_I/v-deo.htmlsi=RBchNK5rGu5tQWWx yep doing something similar in new cops at link above

No worries glad you liked it.

Realtime, tis was Houdini 20 in the old Cops but you can do the same and more with 20.5

Used Houdini cops much like how you might generate textures in substance designer.

Baking textures is generally the way to go for real time. In this case I'm creating textures in Houdini much like you might in designer or Photoshop. The advantage is that I can test texture UVs meshes and animation all in one DCC before going to engine

It's a mix of creating textures in Cops and then creating interesting meshes in sops. You then animate the UVs and the texture flows over the surface. Should see more of this workflow with the new cops on the way.

Thanks. Yeah this one came out very nice

Yes you can get estimated depth using sd. This is largely what I am doing here I guess this is similar to entagma. Not sure about optical flow haven't tried this much though I thought it more useful for getting motion vectors so I guess you need an image sequence. The issue with current AI depth models is that they are mostly lower quality and suffer from significant noise and banding you can clean it somewhat using COPs and then layer the RGB back in. With good inputs you can get some decent results as you can see in parts of this vid. Definitely has some potential.

@@3D_GO yes unfortunately bading is an issue and depht is estimated with AI. I wonder if a mixture both of smart vectors and AI could be the final solution. Also yes, optical flow needs some motion to get the vectors or at least some reference of a distance. Hence my idea of combining both methods might result in a good outcome. Who knows, might be fun to explore

@@borjamassaalmarza3377 possibly I don't really know enough about optical flow. I suspect it is 8 but banding and we will need better depth estimation models at higher bit rates that output to .exr. I guess this will be a matter of time with newer models. The best method I came to currently is to use high quality RGB, generate low quality depth through AI model and then paint back in details from rgb. It's a bit hit and miss for large details for example the nose but is very good for small details such as skin details. Pretty manual currently but given the speed of change in this area photo to high quality 3d is only a matter of time.

My pleasure glad you liked it. There are rumors that COPs will get an update soon. Fingers crossed as textures are generally cheap, high res easy to use and interoperable between apps. The structure is there just needs some dev love to make it really shine.

aw man thanks for the reply. Yeah would love to see an update on that. Seems like the weak side of Houdini

Your very welcome. Glad you like the content. I like exploring lesser used workflows and just hope it proves useful for people.

I did try that. Guess I didn't give it enough shake 😀. Will give it another shot thanks for the tip.

Glad it was helpful!

Glad you like the content. You 100% agree not much out there on COPs at all. Its been a bit overlooked but you can do lots of interesting things with it once you get the hang of it.

@@3D_GO Indeed! Of course, COPs is less comfortable than other specialized texturing apps, but, IMHO, its ability to work directly both with shading and geometry makes it quite unique and potentially very powerful.

@@Rombododecaedron yeah it's got lots of potential. I've been surprised how versatile it is. It's pretty easy to get some awesome shape that would be difficult to generate in Sops.

Thanks appreciate it. Do think that alot of the foundational substance designer ideas can be achieved with Cops and heightfields. While COPs UI can be a bit unintuitive it quite similar to nuke or even shake. It really just need some dev love to make it more stable and faster particularly at higher res.

Thanks appreciate it. Have some more almost ready just need to edit and got side tracked stuffing my face at Xmas lol. Should have another in a few days.

Thanks. Off meta is how I live my life. It takes a bit to get used to but you can do lots in texture that would be quite tricky to do in sops so well worth the initial pain.

Thanks. It's a bit flaky to setup but has lots of potential. Got some more vids on the way that expand on this.

Thanks. I changed my mind on this. I have tried out a lot more textures since this and I have been impressed by what I can create. I think it is very capable once you get used to some of the quirks. I don't think it will be as fast to setup as in substance but as Houdini is so flexible you can create very interesting shapes in sops that I think might be difficult in sd. Its very Houdini the simple setups are hard but the more complex stuff is easier. The last big advantage I think it that Houdini is much broader than sd and the stuff you learn here can be used for FX as well as animation etc.

Should just add I guess it depends on how comfortable you are in Houdini, I guess if your a new user and you are looking for the easiest way to get into procedural texturing I suspect SD might be a better path given the amount of tutorials out there for SD

Thats great! :D Even though simple stuff might be a bit harder. Not having to switch apps is also worth something I think. The other thing I still find hard is when to decide when to model in houdini and when its better to texture certain things/details. Do you have any thoughts?@@3D_GO

@@g8610g in Houdini just because we can copy models everywhere doesn't mean we should. I'm doing more real time stuff these days so I would say only model for larger shapes. If it doesn't break the silhouette don't model it. All the medium to higher level details should be in the textures. We should only really go for procedural stuff for medium distance anyway. The hero stuff should be hand modelled.

No worries glad it helped.

Glad it helped. Probably worth checking out the pinned comment for an alternative way to achieve the same thing using a VOPCOP.

Glad it helped. While COPs is quite powerful it needs an update. The channel copy node is an example of poor UI which could be fixed pretty easily. Copying channels is vital for comp and texturing and I suspect alot of users run into issues with the channel copy. I have a short video with some tips on the channel copy node as I think it probably is useful in some situations but the vopcop method makes more sense to me.

No worries. Glad it helped.

Glad I could help!

Thanks mate.

Coffee and CG all about the dopamine hit

May the 4 to you too

While I think high res 32 bit images will solve it alright i have come away from this solution. Need to have something that will play nice with stable diffusion which is limited res and limited bit depth at least currently.

Pretty flexible once setup. You get good results quickly, but it gets a bit more involved if you want it to look more photoreal. There is loads of fun stuff you can do with heightfields, and Im very excited about plugging in stable diffusion into this workflow, I think its possible to build an interesting workflow around that.

How easy is it to get stable diffusion involved? Would give you image and depth

@@PatrickGavin1 looking at this currently and happy to report it's looking good within certain constraints

Very welcome hope it goes well.

I'm sure it will. my colleague did it in Blender, but he cannot cusomize big swirs and tornado on Jupiter. flowmaps guide are looking good, and i think its the best solution. Houdini rules!

@@ДмитроСказон Flow maps are pretty cool, they get used alot in games for rivers, lava, magic portals etc. I think there are lots of more real time solutions that we can use in Houdini, instead of trying to sim everything.

My pleasure.

Yep should allow for very clever packing throw in a mosaic node and you can pack an animation atlas full of useful info to drive shaders

​@@3D_GO great was also doing some usd solaris to unreal with material x textures works great though crashing when using big geo cache. Hope it will be out of beta soon

@@wewantmoreparty if your using vertex colours on points you will most likely want the labs import attribute node that should do the trick

@@3D_GO Does it work well with animated textures? as flipbooks? Right now using vertex color on my sims, but really wanted to be able to get multiple attributes over

@@wewantmoreparty haven done it for production but I once packed an ocean sim into a texture atlas with displacement normals foam etc. Result was nice lots of fun data for shader to play with. I guess it depends animated textures in flipbooks can get very large very quickly.

Thanks your making me blush! I do appreciate it. To be fair the bar is low with COPs as there is little info out there.

basically want to append/combine black and white masks

too bad man, gonna use HeightField output instead ...

When you say they are all in my red channel do you mean masks from a heightfield? If so you can bring them into cops using a sop import node. Hit the import planes button and if you check the image plane viewer (this is above the viewer and will be set to C by default) in the viewport you should be able to look at each one in the viewport. The masks should show up in the channel copy and you can use the technique in the video to shuffle them. Hope this helps.

@@benox50yes heightfield output is good for direct output disk . If you need to adjust in cops you would need to channel copy. There is another way to do this using a vopcop. If I get time I will do a separate vid on this.

@@3D_GO I can only copy them in the red channel, you cant choose C.G and C.B cause they dont exist with black and white images, and you cant add only C.R etc., you need to add the whole C, so you add the whole "C plane" for adding them in all RGB, but now I cant isolate only one, they always come together. Maybe you can send me a small screen reccord of how, cause its mind boggling

To do this in cops this will require building a system in a vop cop or using vex doable but a bit more complicated. I suspect the easiest thing to do would be to scatter the image in sops using the standard copy sop setup. To get this into cops depends on what you need there. If it's just the shapes easiest would be to use the geo node in cops. If you need the height info displace image in sops then project onto a heightfield. Then use sop import in cops to bring in shape and height. If you need other info say curvature or other attribs there is a lab import attribute cop. Hope this helps maybe I will do a video on this at some point

@@3D_GO thank you for the detailed response. As I understood Geometry node will make something like a shape what is a powerful solution but what if I have some procedural geometry, for example building, and I need to add some graffiti on its walls here and there. How to be? but I guess I can add some randomness to coordinates in cop vop and just place a picture as a layer. Hmmmm but if I need to create 20 grafittis I can make random coords within some interval in for loop inside cop vop. Anyway thank you))

So really I think the technique you are after is texture bombing. This is a relatively old school idea where we randomly project textures onto models. This step is usually done at the shading rendering stage so you adjust as you render. As far as I know it's not implemented in Houdini karma by default. Redshift I believe has an option for it. I think there is an osl shader that could probably be modified to work with karma if that was required.

Now that I think about it there was a talk a few years back from a vfx house I think in Germany where they discussed using this. I will look later and see if I can find it.

ua-cam.com/video/txeyobOnxq4/v-deo.html rise vfx discuss texture bombing setup implementation