Blender 4.1 Split to Instances with Origin

all right we are live hello everyone and welcome to today's live stream where I'm going to be going over the split to instances node but adding the uh instance cell origin points using a technique that arandale uh recently showcased in their latest tutorial so with the split to instances node for those of you who don't know basically takes an object if it has a ton of disconnected pieces like this fractured Cube and it will take those separated pieces or whatever index you have in this case I'll be using the mesh index and it will turn it into separate instances which can be extremely useful but there's a little problem the problem being that these split instances do not have a location output for their origin point which let's just go and demonstrate that so this will be mesh Island Index right here so afterwards if we were to go and rotate these instances we will see that they're all rotating at the exact same Center Point which is not great we want this Pivot Point to be actually where the instance origin is so unfortunately uh there was no implementation of that currently so we have to do it manually with a workaround hopefully in the future uh they will add it but for now workarounds are what we have to do so again Arendale showcased this in their latest tutorial I would highly recommend watching that if you want to get a more in in-depth my bad inter um in-depth example of the effect okay so what we need to do is first uh get the location data or first we need to get the bounding box so what this does is it takes the cells and then it gives the bounding box for each one for each instance this is something that I haven't really gone into too deeply like how the notes handle when there are in es instead of just one object but I need to research that quite a bit more because there's um a good amount of stuff that you can do with it so you'll see that there's the Min and Max bounds here unfortunately these do nothing because this bounding box node acts as if the entire object is doing it it's weird but these ones we can see do not output a field just a sing singular value that is 0 0 which is not useful whatsoever so we need to extract that data manually per in fortunately we can use the um evaluate at index node to get the position at the vertices so position one let's see if we take a look at this one on the uh yeah point should be fine here we can see the location data of the instance or the cube at one point actually wait am I doing this right yeah yeah I am so that's that one and then seven this is the men and then seven will be the max of the cube and we can see the colors reflect uh that so let's go and store this data so let's just go and set this to be zero and then we need to capture this for later so we want this to be a vector on the point domain let's go and collapse that so we want the Min and the max so let's put that to seven bring this over to here so we have both of those stored now how do we transfer that back onto the instance domains well we need to do another hack which Arendale once again uh showcased this in their tutorial where you have to basically make sure that the indices are the same so we have to basically make it so that there's one point per bounding box so that that'll match the one point per instance so that they can match so that we can then transfer the data so let's go and do that so let's just delete geometry and you could just make it so that the index is the input this will make it so that when the index is zero nothing will happen but anything above that uh it will delete it but that's a little it could be a little confusing for people who don't know about the conversions so I'll just put a greater than one or greater than zero will be deleted same thing so we have both of those and actually I should expand these so that we can clearly see it's on the point domain so we capture this beforehand and then uh we let's see is it just me I can't see anything um on my stream preview I can see the noes so everything should be good it might be uh a problem with your setup unfortunately okay so now that we have this we need to store the attribute on the instance domain on the actual instances so let me put two of these here I'll make them named attributes just for convenience so this will be Min and this will be Max so let's do that so now that we have the data here with the bounding box all on one point we can simply sample an index sample index which has this data and then easily transfer it again because the indices match a technique that I need to do more often because apparently like I could probably use convex hole and then get a similar result or we'll see there's a lot of stuff that possibly can happen with this technique but I need to look into it a little bit further so now we have the Min and the Max and they should be transferred just double checking yes that should be good so now to check if this is actually working so this is uh bounds per instance go we go let's just do that and that which should be good there we go so now to check if this is actually working we we can go and feed in uh the Min and the max into the pivot point but we need to mix between actually no let's just check the max here we should see it is working okay it's not working what did I do wrong think I did it correctly point point point point over to there yeah oh wait I forgot to realize okay one thing I forgot realize the instance is after the delete geometry very important there we go so that makes it so that instead of it operating per Point like with the delete node and now makes it so that everything's real so that you can then transfer it in the indic match because this is real geometry real points and then these are instances which are just fancy points in a way so now that we have this we should see that these are now rotating at the max which is nice but if we want this to work with the Min and the max we need to mix between the two so that we get the midpoint there we go so we should see yep that's rotating right there so now um yeah now we could do whatever we want with it which this will pair very well when we eventually are able to make uh a cell fracture node inside of geometry nodes that's performance cuz right now now there are ways to make it but they are the laggiest thing in the known universe but now we have that so that we have the pivot Points I guess I could just store the Pivot Point here and then call it a day but having the bounding box info can be useful in other ways actually no I'm just going to store this as uh origin yeah origin so that we can use this later on so another thing that can come up when I was testing this a little while ago is that if the instances like if you're referencing a collection the instance locations could be different by default because that's how it usually works so you may have to set the actually wait I should probably go and just showcase this so in another collection let's have a few cubes out here in different positions and then let's go and copy this object duplicate that then hide the original there we go so instead of doing this on the split to instances I'm just going to reference this collection need to make sure that this works so we're going to separate the children and see if the rotations are working ah good they are so now um I guess what I wanted to do is to set the position to make sure to reposition all these instances to 0 0 which usually you would just use the reset children button right here but you know that's not really you know let's do it in the fancy way so if we mix between these two scale it by negative 1 and set it to the offset we'll see that this does not work because their origin points are wait hello okay that's working and that's I'm a little confused as to why that's not doing that there we go we're doing that and there reset to there but then again you could just do a vector zero see I think what I wanted to do originally was to oh right make it so that the bounding boxes are centered the Min or the max at one so that maybe you can stack these later on so like this is what I wanted okay so here if the objects are different sizes so here if they're like this we'll see that no matter what shape they are they will be oriented correctly so there and there and then we'll make a big Cube so now these in theory can be stacked with the accumulate Fields node I believe but I don't quite remember how to do that oh well let's try let's see accumulate field I made a tutorial on this a long time ago when it was first uh created but I am somewhat forgetting so I think we need to subtract that to get the actual dimensions and then I may need to multiply that times two then I may need to subtract the leading by the trailing and then put that into [Music] there let's this needs to be on the instance domain see H something's happening but I am really forgetting how to actually do this part of it working like those are stacking but that's because those are both one uh I have forgotten which is very unfortunate oh well I'll need to go and look back at that uh in the future but yeah there we go so in theory you can stack these and it should work pretty well if you know what you're doing which I do not just yet it's been a while since I've done that let's see so that's how you would do it uh it could save you and a pinch but again a big workaround for something that hopefully in the future will just be a part of the split to instances node or better yet the bounding box node just works with instances because as we can see here even though there are instances here it's outputting nothing which seems more like just kind of more like a bug because these bounds work and these are still instances while here it just doesn't output the data for some weird reason so it's good that there's a workaround that we discovered that Arendale discovered but it's unfortunate that we have to do that but anyway so that's that and uh yeah I do not really have too much of a plan after this I guess we could just go and play around with the this so I can see a lot cases where this will be useful like when you're procedurally generating stuff being able to split it after the fact and have the origins and bounding box data can be very useful but I want to see what we can do to push it even further because it's not only the bounding box data that we could do we can use the convex hole and it should work in a similar manner where this is happening per object and I think because all of these are already convex X it doesn't mean anything but that [Music] means let me think about this for a bit it would give us like a lower the lowest poly version of the object that could be used for collisions later on so we could just switch out that with that but we won't get like the we won't get the correct like average points with the convex hole so how would we do that well I think we can use that uh accumul field node and actually use it in the way that I know that I remember okay so let's make another copy of this node group and see if we can get that part functioning so duplicate again hi the old version and then let's go and just delete this then use the convex hole just so that we can get the accurate version and then let's see accumulate field actually wait I should undo this and just keep what we have so this should be good we just need to store the data that we want so I can delete that and that because I only want the origin points so that should work and then this will just be origin there we go bring it over to here this should work I haven't tested this but it should let's go and bring this so this will be uh origin point there we go so this is just uh origin per instance there we go so if we use the accumulate field this should work per instance with all that now want the convex hole oops misspelled it convex there we go let's bring it over to here delete the bounding box so now we can do a vector so we need one for the position data and then one for just the amount of vertices which we could do with an integer set to one so this should work this will be like the attribute statistic sum and mean I guess but working on a per instance basis so actually I should check I should make sure that's working before yep it's work oh very good very very good so that Stacks up all of the positions and then we can use a vector math that should just work I forget if I need to add or subtract one here but it should be fine we'll see there we go now with that we should get the origin points of the convex hole which seems to be working so if you don't want to do the bounding box you can use the convex hole I really should have been using the uh Delete geometry per instance or delete everything except the first point I I should have been doing that for a long time and I'm surprised that I haven't but again special thanks to Arendale for showcasing this technique all right here we go are you in uni doing this stuff uh no I did not go to college at the I graduated in 2020 which as you could imagine uh was not the best time to graduate but I kind of went straight into the um straight into freelance and then working in I let's see 2021 I would say and then it's just gone well from there so no I I do this as a job not in uni unfortunately would be fun though would be fun let's see Al I know I was planning on going to looking into going to college but then I got uh a very good job that uh well some of you probably already know of given my recent posts but yeah um okay so what else could I do with this like in theory I could also what other data could I get CU I could get a fair bit of data like I could store the mass per like an estimate of the mass per cell like the end goal that I want to do with all this is to make rigid body physics with an arbitrary amount of objects because instances and stuff like that but I'm not sure I know how I would do that because this technique could be extremely useful and I think it will it it could lead to that like we have the pivot point of the convex hole which is really really good just not having to realize the instances makes this a ton faster and we don't need to well simulate for each loop with a repeat Zone let's see but what else can be done with this so again all this is just geometry that again while the cells are split I used the cell fracture out on I kind of cheated but then I joined it all together as if uh geometry nodes was able to split all of that in a timely fashion because as we know it can't at least not yet maybe well that people have been working on a bisect node unfortunately like it's been the works for like over a year at this point and I'm not sure if uh the people working on it have time to actually complete it but hopefully eventually we'll get it would be nice in 4.2 but we'll see people have made their own bisc nodes just out of uh desperation and I've seen the setups and they are very complex but I applaud all the people who are able to handle uh that many es and edge cases because it's uh it's complex okay so if I wanted to make this into a rigid body simulation or in a sense where the B the convex holes are accurate but it can detect collision between all of it how would that work cuz I would assume that you would realize the geometry and then each part sees if it's intersecting with another object which that I don't think we can do yet like if I stored every single verticy on the instance domain then then I think we maybe can do that just take the instances see where the geometry is and it already knows its own bounds so it could shift to wherever it wants I don't think so like it feels almost possible almost and yet not so yeah if anyone has any ideas in the in the future let me know CU that would be extremely powerful and way ahead of schedule because again it will be years like I think three years minimum before we get uh Native simulation support in Geo nodes just because we're lacking quite a bit of uh external data and for loops and other stuff but if we can hack it together that means that you know in theory geometry notes could be very powerful in that regard because I've already and maybe I'll showcase this actually one second I will showcase this in the past let me save this as just a test um okay let's open up the rigid body physics system which I believe I put my sim projects file rigid body physics nope that is in 2023 projects there we go so this right here let me turn this back on is a rigid potty a rigid body physics system so here as we can see while this was a test with getting part particles to interact with it which is very funny I didn't expect this to work cuz this is a very hacky setup but we have a string moving the rigid body physics object and then particles uh hitting it which by the way again this is not an accurate rigid body system it's uh it it still has a lot of bus Which technically that's not R well kind of a bug but yeah you can just move all that around and then it'll work no uh unfortunately but this does what regular rigid body physics in blender can no longer do after 2.8 which is having an object that is dynamically moving and having that effect to the rigid body scene cuz right now they're since blender 2.8 there's been a dependency graph change which basically means that if the object is updating its position even if you have the R by settings to deform final all that stuff it just won't update unless the user may manually causes an update by editing a value either in the modifier tab or somewhere else it's really really weird and annoying and unfortunate but it's all we have also it's very funny uh right here this was just a test of the simulate or rigid body state so you can see that the string is affecting that part which is really really cool uh I think I've gone over how this works in the past but it really needs an update cuz uh I made this when I didn't really know too much about all this but basically we have an object we store the velocity and the rotational velocity all that good stuff then I edit the the rotational velocity even though it's not really necessary like that I need to find a better way of doing rotational velocity without just having the axis angle method because that is not great and then uh uh I don't know why I do this oh wait no this applies the velocity and then I store the old position because somehow that can be useful oh because I recalculate the velocity after the fact by taking the current position and the old position and then yeah okay so the way I'm doing this is I'm applying the offsets as if there was no Collision doing the Collision steps and then resolving the velocity after the fact and even the rotational velocity but this is super broken don't okay well it's it's very broken like it works but it's hacky and it's not great but that's that uh but the rigid body collision this needs work because one I didn't know the technique about getting the bounds and all that but actually I think in this case yeah I realized the instances which in this case there's only one it only supports one and then it does all that but in theory I don't need to realize the instance what happens if I why do I join the geometry oh no wait that's for the the line I think I'm not sure anyway does that work nope okay so realizing get is necessary at least for this setup so these are the strings ignore that if you don't want that but this is just the Collision check again checking the normals to see if it's inside the ground object and then uh I take the velocity scale it I'm not entirely sure why I forgot a good amount of this but then I scale by 0.5 because that's why I found works the best moving it half the distance to where it's supposed to be and then I average those values and then I translate the instance based off that and then the rotation is a big pain and there's better ways of doing this but I find the center points which we can find in a much better way now and then we take the offsets with the collision and then we turn that into a rotation with the cross product for the axis and then the r cosine for the distance radian distance between the axes and we put that that there ignore that that doesn't work um a lot of stuff needs to be done then this the live instance just makes sure that the object is currently live which means I can deform and all that good stuff and I also stores oh right because setting the rotation is a little bit yeah no this just makes the instance the current one because when you put an instance into the simulation Zone it's kind of weird so if I didn't do that it would look like this but with the live instance we can see that it's actually working and then I just make draw a line from there to there so that's a brief overview of that still not great as you can clearly see I'm still not sure why a majority it's been a while since I've looked at it and the one part this is an experimental one so it is Messier but yeah work on stuff and yeah like another thing that me and other people have been working on is trying to get proper again rotational physics using inertia tensors but that has been uh very difficult to do that and yeah a lot of stuff I do want to try in other projects that I've really been wanting to uh fix up is the hair physics which I really want to do do but I don't I still feel like I can't quite do it right like I that needs again rotational physics which still is very difficult to do but I will figure it out like in theory it's just like okay use the axis angle wobble both of them but it doesn't feel right to Wobble the axis as a position because it that just doesn't make sense but maybe it does but we'll see we'll see I'll do some more testing but if we can do this then lendini might actually be feasible some of the uh blender developers were uh well Pablo mentioned blendini and uh it would be cool if we ever got to that point but it will be years until native support like without having to do 500 workarounds like we do right now it'll be a long time before we can do that all right and also since we don't have GPU simulation uh all this will be rather slow multi-threaded probably but slow let's see forgot to set up that using a dynamic paint to change the rigid bodies I used it used to work and now it doesn't yep and unfortunately due to those dependency changes I haven't heard a lot about the dynamic DC paint part but um yeah ever since 2.8 all those parts have been at the ends of their lives so there will be probably no updates as I've seen in the uh bug review chats so the only hope is geometry nodes and that hope is very far away so yeah so here's to hoping let's see nice content you got there ah thank you this is again just R&D stuff nothing too special but hopefully one day it could be good because the the goal is just to have one node that is just rigid body physics with all the settings that you want you just PLP it in and you don't need to worry about it again the end go of geometry notes is that you make a node and you never have to look inside of it again which that's again that's the goal [Music] eventually um yeah I don't know what else to stream at this part but maybe I could try out rotational physics maybe maybe see I have a theory on how to speed rendering if we capture a viewport per frame then we can render it at screen shot per rame oh oh that's funny wait screenshot so screen recording the rendering instead of rendering with actually rendering that's an interesting thing that I haven't heard before but um I I would recommend just rendering like well unless you want very horrendous compression I guess you could cuz right now like I guess if I just turned off all the UI and maximized everything then yeah that's really really fast rendering but there's a reason why we render with samples it's because well the default renders look a little splotchy with what given regular e is pretty noisess by default even in real time like most of the time like here there's a little bit of splotchiness but eeve next that thing is extremely splotchy by default and I'm worried that that will not be fixed when it gets released if well if when we'll see but yeah cheers to that day yeah I really hope that uh it Eevee or no uh sorry geometry nodes is a very good base for a lot of procedural stuff it's very convenient to store and edit attributes that's like one of the greatest strengths and and uh and again the integration with the modifier Tab and so on so I hope I hope that we uh that we get to that point um okay what else uh nope that was the wrong [Music] one I was trying to find the bridgid body physics system the robot version ah here we go so this was a test of doing the reg body simulation hello ah there you are rigid body simulation on multiple instances at once which this uses a for each Loop to do so but it's not real for each loop it's a hacky and slow and terrible for each Loop but with the the with what Arendale um showcased hacking that to use the convex hole and then possibly getting that data could possibly fix up a lot of this but right now it's basically doing this for each instance which uh is much slower than it should be as we can see we're getting 9 frames per second in something that in a game engine you could probably do 1,000 of and it would probably run just fine but that's probably also because each instance is looking is looking for its collisions again and again and again while in um a game engine it's probably just evaluating that evaluating that once per frame and then everything just Compares between that uh thing not entirely sure I should research that a bit more okay all right hey I'm new here was just wondering do you do 3D related stuff for work or is it just a hobby uh it's both uh I did it as a hobby for a while uh and then when 2020 hit I decide to start doing it as a career I decide to quote unquote get good and then uh start doing freelance and then from that I got some jobs one of them being well I I worked on the game party animals which came out I believe one year ago maybe two and then uh I also worked on well I did quite a few jobs freelance and not and then I got employment uh maybe more than a year ago which was stable employment yeah two years ago at this point I think and then uh I've also been working on another job which uh I'm very excited about which uh if you want to see what what that is maybe check my Twitter but yeah fun stuff but yes I do do this and not only geometry nodes obviously I do a lot of other stuff but um yeah I do it as a hobby and a job yeah again it it feels like the hobby side has kind of died off with me trying to make this more of a job and then also pushing geometry nodes like in most other programs having rigid biophysics that actually work and you can have it work on you know objects Dynamic objects in your scene and such usually that's implemented by default most people just take it for granted but here in Geometry nodes and again because the old system is broken currently and is at the ends of its life well all the simulation Parts in here we're trying to recreate that in Geometry notes because that is the next way but since we're impatient we're trying to make it right now which it really shouldn't be cuz it's it there's no way on Earth that it'll be faster than the native implementation but we will try because uh it's a challenge and we love a good challenge um oh and by the way this robot is procedural like all all the motion here it's all physic based which is very uh fun which here oh forgot about that I still had a regular Walker in here uh also okay let's just do this so yeah there's the regular Walker and you can see the wires that control where the legs are going oh no I forgot I did that it's cool yeah you have that and then I basically in this part oh I forgot about that oh that inherits the velocity for the rigid bodies so that actually affects the thing I forgot about that so essentially that means if I were to turn off the randomize or the offset like that then they should just fly let me go and exaggerate this they should just fly yeah where they're actually moving huh cool also uh why oh did I hide the original one why did that not appearing [Music] anymore trying to figure out why the original Walker is uh oh that's why okay have that and then it explode there we go so here yeah we can see that the actual velocity of all the parts is actually moving everything I don't think I made the rotational velocity actually work because I still need to figure that out it's beautiful I just wished that uh it actually worked and again no friction because uh I still don't know how to correctly Implement that which I should because I made the particle system and that has physics and such so I should I should Port that into there I watched the new tyo prism trailer and it reminded me how behind blender is in Rich body Sims uh blender is very behind in all physics capacities all of them uh even when Manta Flow came out uh it it has aged pretty quickly yeah all these and again these are probably over 10 years old at this point all of them so it well no I know all these are over 10 years old because they've existed yeah 10 years ago was 2014 actually wait I forget okay maybe not 10 years but uh it's unfortunate that these are now decaying because of the dependency graph changes or the dependency dependency graph change and then the update the fix for it is still like three years away that that's the unfortunate part because they've already been broken for four years now maybe five when did 2.8 come out 2019 or 2020 but yeah it's unfortunate but it'll still be quite a bit longer before we have better implementation and there are add-ons that do better simulations like with smoke and such but um I'm not sure about rigid body physics if there's a better way of doing that but hopefully when the new version comes it'll come with uh GPU acceleration and all that good stuff because it would be nice to have real time uh fluid simulations which the blender devs have stated or Pablo has stated that the new or I forgot who said the new volume notes that were supposed to be in 4.1 but got delayed uh are supposed to be the building blocks for fluid simulations and smoke simulations which I hope that becomes a thing but uh I think it will be a while and then when it does come out at least in the first implementation it will be completely CPU bound which uh means that it will be you know as slow as the old ones but customizable which means that uh mantif flow Shenanigans we can get around and actually Implement particles like the Wake particles uh by ourselves which could be cool but it'll be uh quite a while before that happens all right um other than that any questions because I'm am out of live stream content at the moment let's [Music] see okay so let's see what other things I can open that need work oh I was working on this yesterday but uh I'll be posting this tomorrow and it is the ground fracture which I made this a quite a while ago where the ground fractures and then all this happens you know the trees wobbling and all that which believe yeah the video should still be up on how I kind of made it and such but I just updated it with actual assets like actual pre assets the ground and such so looks a bit better so I'm hoping one day I can really flush this out if I had time but for the moment it's it's okay how was your experience with getting freelance gigs at the beginning uh it was good I had a pretty good social media following so people would usually come to me for their freelance gigs which I that's a very different thing than what most people do but since I had that following uh people usually came to me rather than me going to them which is not the standard experience from what I've heard from other people but that it went well uh I would just yeah and at that time I did not have to support myself financially so it did make a good amount of money not enough to survive if you're having to support yourself financially but it did work uh but I would recommend if you're looking to get into 3D work try to find stable employment salary based because then you just don't have to worry about finances because with freelance you have to wonder oh am I going to make rent for the month and so on and then trying to get job well trying to get gigs and such but with actual stable Employments uh it you don't have to worry about it so I would recommend that freelance is very good starting out it'll teach you how to deal uh with client requests because working professionally it's the workflow is pretty similar just reliable employment is oh you have the same client for a long time if that makes sense it's similar and the people in the industry know what to expect because they hired you to do a certain thing and you do that thing while with freelance the people who hire you sometimes don't know the limits of like the 3D software and what you can do what your special well what your special specialty sorry what your specialty is which uh at some points they can request oh could you add like this fancy thing in the background and you're like uh unfortunately no and that kind of feels bad because it's like oh I should be able to make this entire scene freelance with freelance a lot you have to be the entire pipeline most of the time if someone's like hey I want the scene you have to texture it render it uh put in all the Sims if there are Sims and all that but with uh working in a company or something like that you are a part of the company you are a part of the workflow you don't have to do everything which is really takes the stress off CU they hired you to do a thing and you do that thing and if you do that thing well it's good so freelance I would recommend it just make sure that uh you pick your clients well that they know what you're cap what you're capable of uh and in what time frame and also uh make sure that uh you get paid well for for your time because with freelance again you don't only you need to take into account that the time you spend doing your business isn't only when creating the content like you need to charge for the time that you spend searching for job well no not charge the clients but over like the span of a month you have to be like okay how much time did I spend searching for clients and getting the work versus actually doing the work and then factor that into how much should I charge per hour when actually doing the work and there there there's a lot of stuff to deal with when um doing freelance a lot to think about which can be a little bit stressful at times Well in regular employment a lot of that is handled for you in a sense in a sense but yeah I still do freelance every now and then if someone that I know needs uh help with certain stuff and uh yeah it's still good yeah that's super good insight thanks a lot pal hey I'm happy I could give some advice I do work at a studio but the economy is not great so I'm trying to get some extra C some extras yeah um yeah like you at the moment like I still have employment but when freelance gigs come around sometimes I pick them up if it's like oh wait I could treat myself fun and sometimes like especially with geometry notes it's a very Niche thing that not too many people do so if people need help with the geometry nodes task there's only a handful of people that can do it to the level I can so it's good to help out when uh when people need your very very Niche skills which I guess that's a good a good thing like if you pick a very Niche thing to be skilled at that means that when people need someone to do it it's usually you like well the competitiveness is a lot l less because uh there's less people in it like Geon noes there at the moment as we've seen even in this live stream there are a lot of workarounds that need to be done uh but if you can get around those workarounds like I kind of fell into that I'm not sure if I were to restart all my knowledge from scratch I think I probably would pick Houdini just because it's more powerful but since I know Geon noes and the environment very well I can use that to my advantage like everything in here is doable in Houdini Houdini is the best in the business and always will be like this cannot do anything with heck it can't even get the uh active camera data outside the location rotation and scale you just can't get like the fov none of that unless you use a driver but that's hacky and not great which again is why I'm trying to get that VI layer attribute node in but we I need more time to study C++ um but yeah that's that's that but yeah what once geometry nodes is steadily improving and uh I hope that we get to the point where you know we can make the rigid body physics and all that the hair simulations ourselves and that there's less uh roadblocks when it comes to that the biggest one the biggest road block I would say in terms of just doing effects and not in terms of like a pipeline is the 4 Loop probably which the blender Dev said it during the blender conference I I heard from other people I wasn't there at the blender conference this year maybe next year or last year maybe this year I'll I'll be there but the 4 Loop would solve a lot of the problems that people brought up during that and uh it is planned but at the moment there is no development M on it there was a mockup made but nothing beyond that which I think is because when they created geometry nodes originally they kind of thought that every node would be its own for each Loop like set position is a for each Loop and all that and a lot of the nodes they're updating to make it so that you don't need a 4 each Loop to do stuff like the fill curve node used to not have this group ID input but now that it does it has a 4 each uh I'm assuming it's doing it for each Loop per curve now rather than just doing it on all the geometry at once so they're making it so more nodes are for each Loops themselves but it's still uh without an actual Forge Loop things are still rather lacking is it possible to do Motion Graphics in c4d like c4d and blender could you I would say you could but it won't be as convenient if you know c4d then use c4d but uh if you know blender use blender it's just about which one you need I can't get used to the new auto smooth the new auto smooth is better in so many ways uh I've heard even in my own comments section uh people being like that they don't like the new Auto smooth like it is a change of pace but it is just better across the board like here you can actually see oh which parts are well actually here let me go and do this so here shape Smooth by angle you can actually see now which parts are sharp with that it doesn't update automatically unless you have the modifier so let's go and put in the smooth bi angle modifier and now it's actually oh I can actually tell what's happening Auto smooth or smooth bi angle is now where it was always supposed to be which is in the modifier tab it really should not have been in this part at all well I guess in in some I I can kind of see what they were going for with having that in there but it was so unpredictable having Auto smooth it it was well it wasn't as reliable as just having a modifier where you can you know you apply it and then this will not change when you move it cuz before or how would you apply Auto smooth you couldn't really I guess maybe if you exported and imported it it could apply it but now it's where it's supposed to be in the modifier tab which is very nice I don't know it just gives me results different than what I was used to results should be exactly the same from what I've seen in the dev chats I think they said that that it's supposed to be the same uh I mean there is the ignore sharpness thing but I think the default value is just it should be the same but now it's just a lot more predictable and usable with other workflows which I've seen people add-on creators that are very very happy about that change and also I'm not too sure about that I've heard that in the actually no I don't know enough about that um but now it should be faster as well given that uh given that the normal calculation inside of Geon noes is faster so like it does seem like I don't remember results like that when I would use Auto smooth and then bring it down so maybe it is doing it per Edge but that is better I forget back in the day I don't think it would do that I think it would be per Vertex or kind of like that but now it's happening with that which is actually really cool so if if that's what you mean by kind of unexpected results I I can see that because now it's a bit more specified but that means it is more accurate me try oh okay interesting yeah I don't think I've ever seen uh Autos smooth do that to a UV sphere before but that is interesting all right we're coming up on the 1 hour mark so I think we can uh wrap it up also the new modifier tab I know people dislike the new modifier well a few people well a good amount of people dislike the new modifier Tab and I do understand that in terms of UI like it's more clicks in order to do stuff but I think well unless you just type like there is just the oh you can just type in subdivision surface and then you get that well it is more organized but at the cost of not everything is two clicks away now it's kind of like that and then you have to search for that and that or you could just search it normally so functionally it is better and more organized I can see how people don't really like the change all too much especially since the constraint thing is still the old one I don't think that this one will change anytime soon because you know there's not too many where you have to compress it [Music] yeah they added the sheen bstf node I thought that was a thing I thought uh Sheen bsdf was a thing for for a long time huh maybe that's a 4.2 thing cuz I know there's Sheen in other parts but oh this part yeah there Sheen in here but I don't see it as a node I haven't really used this too much only in testing yeah um that is what is happening in blender this week uh I'm looking forward to the Google summer of code projects which quite a few people are uh talking about adding some nodes there are a few nodes I'm interested in like the sound nodes and such but um with the sound nodes the I saw that the original proposal was from March 2023 which means that it's been over a year since then so I hope it comes around this time but again with uh you know don't hold your breath until it's actually inside of uh the daily builds and even then there's a chance that gets removed like simulation nodes but um yeah okay so thank you all for watching I hope you've had a fun time I hope that the um instance uh origin Point SL bounds data was useful thank you very much sarale for showcasing that and I hope that I can somehow hack it together to the point where my rigid biophysics system can work with that because I feel like it's possible I just need to see what the limits are when the geometry nodes are happening on the instance level because like set position that works on the instance level not on the Point level when there are instances so maybe there's a way to work around that but I'm actually not well actually wait I need I I am going to test that right now uh sorry one second everyone so if this is two instance or geometry to instance I'm also kind of oh no that makes sense so here and then we have set position we set this be random this will not uh modify the vertices of this per one but if we have a stored named attribute node if we put in the position and connect that to the output it does work okay okay so that's how you set the position again the stored names attribute node is the best node in Geometry nodes it's just most of the other nodes but better I'm I really love it so the stuff that the regular set position node doesn't do this can so and this can be per instance so it is superior and I need to remember if this is true but I think I saw somewhere that there was the consideration of removing default attributes from attributes that have their own inputs and set position and nodes like if there's an input which is position and then a no that's set position I've I do need to verify this so take this with a grain of salt I've heard that they might remove those from the store name attribute node like it being able to see and read it if it pre exists like an actual node exists for it but I think this is a really good argument for not doing that because this again unless you want to put a domain uh level a donate domain en on all the notes that can do this uh you know generalized node is better but that's good to know I need to take advantage of that again I haven't done enough with working like figuring out how uh messing with geometry works on a per instance level because I could abuse that to make that a kind of for each Loop cuz I assume that this works for each instance and is multi-threaded okay I think you can apply Autos smooth under data geometry data add custom split normal data yeah custom split normal data like that usually shows up when um and I think eventually that this will probably be removed when geometry nodes can set normal data because I think at the moment normals are kind of a weird um kind of field like like it's not a Anonymous attribute I believe they call them kind of like how actually I forgot what that was I guess kind of like how vertex colors were before they changed it but I think that this will probably be removed sometime in the future because they do plan on adding custom normal supports in um geometry now modes so you can set the normal kind of like how you can set the curve normal they reference that when they added in the custom curve normals So eventually we'll get there just takes a lot of time all right thank you all for watching I hope you all enjoyed and I will see you all next time have a good one

Channel: Cartesian Caramel

Views: 2,681

Keywords:

Id: 1HKFSqdLJEU

Length: 62min 49sec (3769 seconds)

Published: Mon Apr 01 2024