❤ Community Sponsor: Spencer Magnusson 🔨 Optimize your scenes with the nView addon: blendermarket.com/products/nview-v3?ref=94 🔨 Get the Light Painter addon for free: github.com/semagnum/light-painter
I guess the easiest was to explain SDF modeling in comparison to poly modelling is : With SDF you model the volume while with poly modeling you model the surface
It might be easier to use a comparison thats already understood such as traditional polygonal rendering vs raytracing - SDF is like ray tracing where objects can be described as functions - you can query a sphere function with coords XYZ and get a true/false for inside or outside the sphere, query that function for every co-ordinate in 3d space and you'll get the sphere. Combine thousands of these functions that both add and subtract from space and you can "describe" complex objects with infinite detail and no set resolution limitations
Having worked both in CAD parametric modelling and in low-poly game assets for over 25 years it still astounds me how many artists still fail to get their head around this subject, and I think you've done a pretty good job of explaining it. I've always found a lot of poly modellers scorn my traditional spline based approach to modelling, purely because they've only been taught sub-D methods. Many fail to realise it's because you can guarantee clients will change their mind and so the concept/parametric/virtual (call it what you like) model is usually almost always easier to go back to. Doing art for fun, and doing art for commercial clients are two different worlds.
I used POV-Ray a lot back in the day, I kinda feel nostalgic for it tbh. I would like a text based modeller to play with but with modern render times :) lol
@@eternalreturnal sure, I might have gotten the years a bit wrong. 3dsmax was also around I believe. But mesh renderes of the time needed super expensive hardware to not be trash. And the software was not available at any reasonable price. POV-Ray was free 😁
So excited to see SDF modeling in Blender 🎂I've been following John Kaz myself and the add-on; it'll be amazing. Great explanation, Curtis! Good balance of the high- and low-level.
Respect to this man. Not getting any corporate sponsorships for sake of our community. Yet gets individuals from the community. This surely limits your growth, but you are the only content creator i know i can %100 rely. I press F.
Metaballs are a special case of SDFs. They also have an exponential falloff instead of a linear one, so they are not strictly DISTANCE fields, but the same is true for all SDFs once you start messing with them, like bending or twisting or even scaling.
Mediamolecule made this years ago with dreams on the ps4. Been using it myself and its realy fun and intuitive. Happy to see that blender is going to implement something similar because its not possible to export your sculpts from dreams
Probably the easiest comparison for SDF is the difference between vector based and pixel based images. One is described by Vektors, which means you can zoom infintly into it without loosing detail. While pixel based images are limited to the resolution of pixels you choose. Just that only with polygons compared to SDF.
Hey, about time Blender got this. I've been volume modeling in C4D for years, it's an exceptionally artist-friendly way to build complex shapes (especially organic, but this is a great example of how it's useful for hard-surface, too) without needing to give a second thought to topology, intersecting geometry, or any of the other pitfalls of boolean modeling.
Another point.. I think historically the reason we used polygon modelling and why it was convenient, was because we were able to write algorithms that could run relatively quickly to draw a 3d polygon in space, projected to a 2d viewport, and fill it. Because this was fairly efficient, hardware acceleration was quickly devised for it and that became the standard way of rendering CGI (remember the old Silicon Graphics machines?). From there the industry poured all their resources to improve the fidelity of the polygon rendering (to handle intersections, render speed, transparency and the like) and then projecting 2d textures onto the polygons to save having to render more detailed polygon meshes. Projecting a 2D texture skewed to match the 2d projection of a polygon was even faster than that many more polygons. That all led to the market being laser focused on polygon modeling, despite all the compromises and drawbacks. Some of us that came from a more engineering or scientific background were still trying to use procedural things like NURBS and Raytracing, but an industry that was all about just churning out images, high fidelity was not important, speed was. It's only now that computing power has gotten so far that we're focusing back on fidelity of the render with raytracing, and now fidelity of the models, with these new SDF modeling concepts, and others like them. Now that being able to rapidly generate high fidelity content is the limitation, those are the problems that industry is focused on, so that's why we're only now seeing so much drive for things like hardware raytracing, procedural generators being built into things like Houdini, Blender, and others.
Great video. I've been using SDF modelling for the past few years through a PS4 game called Dreams, and I'm excited to finally see SDF implemented in Blender as well. This will be great for hard surfaces, as shown in the video, but I think SDF really shines when used to create organic shapes.
Polygons - a composition of straight lines. Because the line is a simple (linear) function, that's easier to compute any position on it. NURBs/splines/curves can have multiple control points and weights that affect a position on the line, so it's more demanding on computing a position.
Wow. I can not say anything about the SDF Modelling work as I have been a lighting Artist for many years, This free addon will be a game changer for me. Thank you.
You could cache various "real" versions of that "virtualized" object on first load. For games that would mean that designers/modellers can focus on the shape itself and the engine can bake/cache multiple versions of this object with vertices on load or even at compile time already. If it can be optimized enough to work on a larger scale of objects it would basically mean free and far finer control and availability of LODs at any given point in time. Combine that with lazy streaming of assets and it might not even impact performance that much with a massive gain in more consistent but variable fidelity.
10:34 - Subdiv modeling isn't a lie, it just isn't implemented the way Ed designed it (at Pixar or before?). Real subdivs automatically subdivide to the sub-pixel level, so no matter how far you zoom in, you always get a mathematically perfect view down to the resolving power of your pixels (and yes, SDFs have the same pixel limitation). The fact that devs write faulty algorithms which stop the subdividing prematurely isn't the fault of subdivs, it's the fault of the devs. The same thing would happen if devs implemented faulty SDF code or faulty NURBS/spline code.
🎂Super excited to try out SDF when it's released and I have some free time. I think it's promising. Having tried 3d coat, plasticity/MOI/F360, substance modeler for hard surface concepting - I find nurbs surface modeling to be too restrictive, voxels are great, but remeshing tends to get slow. Sculpting while freeform is difficult to achieve anything non destructive precise, bevels are particularly tricky and time consuming. I do think the best approach is a lossless component editor combined with a lossy polymodeling workflow. In many audio editing DAWs there's a concept of "freezing" that prints to a lossy format. I'd love to see a UI that allows you to freeze/unfreeze the SDF generated components via some sort of LOD or proxy system. Didn't expect to see a dragon cameo in this one - thanks for the shoutouts!
This is a game changer! I switch between parametric, Boolean and polygonal modelling constantly and this appears to be a far more natural approach. The lack of hard edges and conversion to mesh is not as crisp but I'm sure there will be workarounds. I'm excited to see the progress.
Wonderful I was meddling with plasticity and loved it only getting frustrating with not have the same control Im use to with hardops an blender in general. this could be the next evolution. thank you for this awesome info.
I thought your sponsorship explanation video was a cool idea, but seeing it in practice is even better. I’m getting Light Painter today and bookmarking the other add-on
I have been working with vector based software to create 2D stuff, I really wanted to jump to 3D modeling but I couldn't get my head around why I cannot just blend objects and curves like I used to do it with vectors... Ha! so this is why! now I can create with my same mindset but in 3D!!! Thank you!
I think you should also review "plasticity" because it is also a very new and exciting software which is targeted for hard surface and gives you very high level of flexibility and even with very good precision.
This is incredible. My modeling background (hobbyist.. not pro at all) is moslty Solidworks/Fusion 360. Im very accustomed to CAD style modeling. Ive had trouble with poly modeling, but i absolutely see the power of Blender. This could be a great gateway ....
I'd think abstract is a better word than virtualized. It's a representation of the mesh before it's turned into a concrete view. Probably stored as some kind of graph data structure under the hood. But that's my best guess.
SDF modeling reminds me of how the game dreams for PS4/ PS5 approaches he modeling process. I think that game / creation enviroment acheves a similar approach with some sort of point cloud modeling process but none the less anything that is as fast and versitile and lets you create actual geometry in the end is very exciting. Thanks for sharing as always
Might be some misconceptions about procedural materials. They are not in fact, 2d materials wrapped onto a surface, and do not even require UV unwrapping. In fact, if you make a procedural material with some noise to it, and slowly animate the surface shape of an object while keeping the material static with the object's origin, you can see how the material indeed permeates through the object continuously. The limitation is actually on the rendering engine's side of how it computes the material. If I recall, even Cycles only computes the texture upon intersection of the surface, even if the surface is transparent. I thought there was some improvements coming to this, and volumetrics can provide a solution. I was loathed to ever use UV textures because I felt they were just wrong. However, for simple things like wear and tear, I eventually learned that a simple rough texture map was far easier and quicker than trying to make a realistic procedural material. I probably could have sped it up with using addons, but I tend to always do things the hard way in an effort to learn how they work, I guess.
Your component/ modelling pipeline description sounds perfectly reasonable to me. After some time with Blender I still had to get used to that "virtualised..is it real or not" kinda invisible realm thing you described. It would be good to have a way of formalising that workspace.
The idea of something being virtualized is interesting because it's a concept that moves with the frame of perspective. For example, if the real world is our working environment, then everything on the computer is technically 'virtualized', and the word loses meaning when comparing digital modeling techniques (it doesn't matter whether something is polygonal or not). But since here we are implicitly considering our working environment to be the 3D view in Blender, using 'virtualized' to compare modeling methods suddenly makes a lot of sense, because it answers whether we have polygons readily-available to manipulate. In the case of a virtualized technique, the answer is usually no. In the case of a non-virtualized technique, the answer is yes.
@@CurtisHolt Interesting. Perhaps, as you say, when a broader range of techniques are in demand in the near future, the necessity of separating composition from casting (to use physical terminology) will be more readily evident. Makes sense to me. Cool vid
Yep (as far as I know), there are a variety of new (and old) softwares making use of SDF techniques in various capacities. It looks like interest has picked up in recent years.
Seems like a boolean function that joins objects and uses proximity and maybe intersection edges to apply a 3D gaussian blur/average to the vertices. Very impressive and very fast. It would be interesting to be able to choos3 the time of transition/blur, as well as the radius.
Really interesting. Reminds me of Real 3D on the Amiga in the ol days. I agree that it would be ideal for 3d printing, as we seem to be dealing with solid shapes, and there is less chance of missing or dupe poly's. It also looks like it has built in 'cut away' views to examine internal structures. I’ve been looking into CAD apps, and it looks like this does things in a similar way. Would be great with ‘constraints’ applied. Be amazing if controlled with Blender animation. Bullet holes strafing a fuselage. Towers growing out of scenery, realtime motion gfx cut reveals of illustrations, bugs running around under the skin. Spacial textures would be great for cutaways of concrete slabs and building frames. Interesting to see if the textures would bleed within the fillets applied?
Hi all, have you tried Clavicula which is free ? This is my personal modeling room and blender is the mesh "operator", my two devices to process. Thanks for talking about voxel modeling, nurbs modelling, cad modeling, polygonal modeling. And the ultimate SDF modeling. Which is of course more flexible. Only one thing is missing : it is shader based and the colors are not preserved when meshing is process. Am I wrong ?
Technically metaballs are SDF spheres. The fact that they are triangulated in blender, just means that internally are not rendered using a raymarching technique, but are rasterized to polygons on the fly to use the same gpu pipeline as meshes.
Just say this on my recommended and it's wild! Thank you for sharing this! I have one question though: what's your thoughts on SDF modeling for organic characters? All the examples I saw look like they are for hard surface, but it seems like this method could make molding a character pretty intuitive.
The analogy of a pixel-based image from a vector image would probably fit. If in 3D, the polygon based modeling would be pixel-based and then the parametric based modeling NURBS is the vector based. But these are for the surface or faces of the object.
There is a game called 'world of goo' which uses fuild joined circles to create web type playfeilds. This could be applied here to give some more organic type structures.
Quad remeshers are making big strides. Exo's quad remesh is still great although I don't know when it was last updated. Seems like a perfect use for it.
As someone coming to 3D from a photography background this sounds very similar to the dynamic between PNG & JPG. One gives you infinite detail via mathematical functions that is real handy for clean cut shapes & anything that you want to scale up or zoom in on. While the other is a compromise that is great at capturing the real world quickly and efficiently, but it does this at the expense of loss and general messiness, especially if scaled up or down (pixelation being the equivalent of the subsurf scaling issue you discussed). Both have their completely separate uses, which is why one format does not dominate and most people develop workflows around knowing when to jump from one to the other - so couldn't agree with you more about your workflow discussion.
Component to Model is my workflow when working with Rhinoceros NURBS. After construction, I tessellate to a poly mesh and then export to Blender, Maya, etc. The bottleneck is cleaning parametric ("naked") edges first before export, since they are mathematical approximations and not explicit as in poly modeling.
Oh wow I'm excited that it crossed your mind because I wouldn't have thought that anyone would remember (or care). Yes! Getting used to a swivel chair again has been weird. I've learnt so much about neural adaptivitiy over the last few months. One thing I didn't expect was that neural expectation for balance is much more context-sensitive than I thought, meaning that you might experience different symptoms simply based on where you are, what you are looking at, even if the exact same motions are affecting your body. Control is also super important, if someone wobbles my chair, I will get a strange balance disconnect (signal mismatch) almost immediately and feel a vestibular migraine event coming on, but if I make the same motions myself, I am fine. My brain knows exacly which motions are coming, so neural expectation is not mismatched with mechanical signals. Sometimes if I'm eating at a table with people, I have to move away and eat elsewhere because they accidentally create too much movement. With that in mind, I wanted to test a swivel chair again because it would be more convenient for the space (and I felt like it would be a personal achievement if I could do it). The first few days were pretty bad, not gonna lie (especially the evenings when I had my lights off, making it harder to keep the room as a frame of reference when looking at the screen). But it seems that by persisting, I was effectively training my brain to learn how much force is equivalent to movement and get used to the control of the chair, specifically in relation to the screen / desk. Fast forwad to now, I can flop around the place and there are no side effects, however: I feel differently when facing the other tables. That's the weird thing about the contextual brain training. I feel fine at the main desk moving around on a swivel chair, but stranger on the others, so I can only stay at those for so long. There's still a lot of work to do (I don't think it's possible for me to ever feel perfect). In this rental, there's a part of the kitchen with an angled corner against a diagonal wall, and when I approach it, sometimes I get weird balance signals and have to back off. My nan has nice bar stools in her apartment, but they curve down a bit at the front, making my brain get confused and think that the world is slightly tilted forward. So there are still these things I need to be aware of. You'd think I wouldn't be able to sit or go anywhere but... Cars, trains, planes are fine. It's entirely a contextual tolerance (I still get motion sick in vehicles though, which is a very different feeling to my regular strange signals). I've also made some good strides in finding dietary things which make me more sensitive to the sensations. Some of it is pretty obvious: alcohol is the worst. It affects the interpretation of the vestibular nerve, so it throws expectations way off-base. Caffeine is also inconvenient but not as bad for that, however after testing, caffeine spikes did trigger new BPPV episodes. They were all anterior canal episodes (weirdly). Not knowing why that happens keeps me testing new things. Additionally, my resistance to vestibular migraines was good before moving to this rental (after coming off lansoprazole), my theory was that I was absorbing more minerals. However after one of the new anterior episodes, I had an awful vestibular migraine event, it felt like I had lost the resistance to them again. That didn't make sense until I considered that I had been much more active here - walking, sweating in the sun, climbing up and down hills to upload content. With that in mind, I've been testing electrolyte supplementation (first in isolation, now in parallel to caffine), to see if replacing them makes a difference. It's too early to tell, but I'm hopeful.
@@CurtisHolt it's crazy to me how your knowledge of what surrounds you and your brain interpretation are two separate things, I'm glad you're still pushing, and found ways to make it happen less frequently. Ya I remember it and cared I've watched the whole video and found the subject quite interesting since I haven't heard about it before !
another old 3d program used pointy ended sausage shapes to simulate muscles (like the still suits on David Lynches Dune), which were stacked together like muscle fiber bunches, and then resolved an organic shape when the meta ball like smoothing smooshed them all together.
@@ViktorasMakauskas Humm, I posted a link but it may have gone off for verification by the channel. I think it was metareyes-plug-in for 3dsmax. I used to ponder 'back in the day' that if one defined the 'muscles' that formed the skin with the meta smoothing, that you could then link the 'musles' to the ik and get the 'muscles' to reshape and thus change the skin topography???
the only potential problem i see with this video is that you describe SDFs as being "abstract" while polymeshes being "real" - but both are just different ways of expressing surfaces/volumes. This is similar to saying SVGs are "abstract" because there are no actual pixels in the file, but we all know that they are "real" and sometimes even better for some applications. SDFs are not converted to meshes or point clouds during rendering (unlike SVGs are turned into raster images for rendering) and hence, can be better than poly meshes in some areas - such as CAD / 3d printing / CNC and can have better performance and resolution in raytracing (raymarching).
SDFs are turned into raster images for rendering just like SVGs. All 3 are the same in that regard - polygons, SDFs and SVGs are just descriptions of functions that, at render time, are used to compute the position/color of a pixel. So I would say all of them are just as abstract AND real as each other. Unless you define real as a bitmap.
All the native SDF tools need at this point is a straightforward way to subtract volumes, then all bets are off. Geo nodes SDFs are OP and nobody knows bout it. Thx for the vid
I'd imagine you could use discontinuities, areas of high curvature, etc. in the SDF field as a way to place geometry and topology algorithmically, where they'd follow the SDF exactly.
how do i buy it (SDF add-on), would happily pay like $25-60 on Gumroad even for like alpha access + perpetual key; dont care if it crashes or blows up. I work in digital avatars/3d printing/organic surf and need something like this right now.
SDF was around first with #MadeInDreams #DreamsPS4 since 2013 I personally am making short animations with it! then there are other SDF apps like #Unbound #MadeInUnbound game engine and #MagicaCSG and #Womp3D and #Calvicula and now we are getting it with Blender with #ConjureSDF and some inside Blender tools too! glad to see it's happening for Blender! I personably made some tutorials for Unbound, models with it and short animation with Dreams! funny thing I didn't use it for hard surface at all! I used them for making characters and curved objects! you also can texture paint with it!
No mention of Adobe Medium or Adobe Modeler - pure technical marvel of SDF sculpting and technology.. not to mention the Move and Warp tool which no other app can reproduce - 6Dof move in VR
there is another issue in that its very expensive to continuously calculate all these things in real time. if you could take that final model and then turn it into a standard polygon model that would be good. One problem i had with node based procedural modeling with blender was the fact that it was annoying to convert to real mesh that doesnt melt my gpu.
Hi, I see the git hub repo for the light paint addon. It contains a bunch of random files and nothing that says 'instal me' what do i need to actually do to get it to work ?
0:15 Oh, Metaballs get improved? Some time ago I tried to use Metaballs to create effects like these in order to make multiple slime people turn into one.
❤ Community Sponsor: Spencer Magnusson
🔨 Optimize your scenes with the nView addon: blendermarket.com/products/nview-v3?ref=94
🔨 Get the Light Painter addon for free: github.com/semagnum/light-painter
I guess the easiest was to explain SDF modeling in comparison to poly modelling is : With SDF you model the volume while with poly modeling you model the surface
Oh I see, I have experience with solid modeling with Solidworks.
Basically modeling in dreams by media molecule their modeling system is entirely based of this i think
@@thesplynter8426Yup, voxel based sdfs in way
@@scaramouche7759 No, NURBS is still surface modeling. When you create a closed surface, the program just treats it as a solid.
It might be easier to use a comparison thats already understood such as traditional polygonal rendering vs raytracing - SDF is like ray tracing where objects can be described as functions - you can query a sphere function with coords XYZ and get a true/false for inside or outside the sphere, query that function for every co-ordinate in 3d space and you'll get the sphere. Combine thousands of these functions that both add and subtract from space and you can "describe" complex objects with infinite detail and no set resolution limitations
Having worked both in CAD parametric modelling and in low-poly game assets for over 25 years it still astounds me how many artists still fail to get their head around this subject, and I think you've done a pretty good job of explaining it. I've always found a lot of poly modellers scorn my traditional spline based approach to modelling, purely because they've only been taught sub-D methods. Many fail to realise it's because you can guarantee clients will change their mind and so the concept/parametric/virtual (call it what you like) model is usually almost always easier to go back to. Doing art for fun, and doing art for commercial clients are two different worlds.
Fun stuff, this is how everybody used to do 3D 20-25+ years ago with programs like POV-Ray. It was all mathematical programming of basic shapes.
I used POV-Ray a lot back in the day, I kinda feel nostalgic for it tbh. I would like a text based modeller to play with but with modern render times :) lol
@@ianstubbington2334 I still use openSCAD for 3d printing models, it exports fine to blender 🙂
25 years ago there already were 3d modelling software... Hell,Maya is almost 25.
@@eternalreturnal sure, I might have gotten the years a bit wrong. 3dsmax was also around I believe. But mesh renderes of the time needed super expensive hardware to not be trash. And the software was not available at any reasonable price. POV-Ray was free 😁
It's very similar to old school metaball or NURBs workflows, but with some distinct advantages depending on the refinement tools available!
So excited to see SDF modeling in Blender 🎂I've been following John Kaz myself and the add-on; it'll be amazing.
Great explanation, Curtis! Good balance of the high- and low-level.
Congrats on the sponsorship! 🥳
@@Wendy3Dimensional thank you! ☺
Thank you for light-painter
@@Igoreshkin you're welcome, I hope you find it useful 😊
Never heard of Light Painter before. Have to say it looks really cool. Nice work.
Always a pleasure to watch your content. 🎂
Wow, thank you, that’s super generous! I’m really glad you enjoy the content. 🥳
Respect to this man. Not getting any corporate sponsorships for sake of our community. Yet gets individuals from the community. This surely limits your growth, but you are the only content creator i know i can %100 rely. I press F.
Best explanation I've seen so far, thank you
First
Excellent video curtis... You explained it very clearly without getting too technical : )
🎂
I'm finding myself starting a lot of projects in Plasticity and bringing them into blender. Just like the component room to modeling room! So rad!
Plasticity is black magic
Metaballs are a special case of SDFs. They also have an exponential falloff instead of a linear one, so they are not strictly DISTANCE fields, but the same is true for all SDFs once you start messing with them, like bending or twisting or even scaling.
Mediamolecule made this years ago with dreams on the ps4. Been using it myself and its realy fun and intuitive. Happy to see that blender is going to implement something similar because its not possible to export your sculpts from dreams
I wish they made a pc app someday
Google "POV-Ray" ;-)
Probably the easiest comparison for SDF is the difference between vector based and pixel based images. One is described by Vektors, which means you can zoom infintly into it without loosing detail. While pixel based images are limited to the resolution of pixels you choose.
Just that only with polygons compared to SDF.
Excited to see the sponsorships start to be utilized on the channel! 🎂👏👏👏
Hey, about time Blender got this. I've been volume modeling in C4D for years, it's an exceptionally artist-friendly way to build complex shapes (especially organic, but this is a great example of how it's useful for hard-surface, too) without needing to give a second thought to topology, intersecting geometry, or any of the other pitfalls of boolean modeling.
Volume builder is so insanely helpful in so many ways
Another point.. I think historically the reason we used polygon modelling and why it was convenient, was because we were able to write algorithms that could run relatively quickly to draw a 3d polygon in space, projected to a 2d viewport, and fill it. Because this was fairly efficient, hardware acceleration was quickly devised for it and that became the standard way of rendering CGI (remember the old Silicon Graphics machines?). From there the industry poured all their resources to improve the fidelity of the polygon rendering (to handle intersections, render speed, transparency and the like) and then projecting 2d textures onto the polygons to save having to render more detailed polygon meshes. Projecting a 2D texture skewed to match the 2d projection of a polygon was even faster than that many more polygons.
That all led to the market being laser focused on polygon modeling, despite all the compromises and drawbacks. Some of us that came from a more engineering or scientific background were still trying to use procedural things like NURBS and Raytracing, but an industry that was all about just churning out images, high fidelity was not important, speed was. It's only now that computing power has gotten so far that we're focusing back on fidelity of the render with raytracing, and now fidelity of the models, with these new SDF modeling concepts, and others like them. Now that being able to rapidly generate high fidelity content is the limitation, those are the problems that industry is focused on, so that's why we're only now seeing so much drive for things like hardware raytracing, procedural generators being built into things like Houdini, Blender, and others.
Great video. I've been using SDF modelling for the past few years through a PS4 game called Dreams, and I'm excited to finally see SDF implemented in Blender as well. This will be great for hard surfaces, as shown in the video, but I think SDF really shines when used to create organic shapes.
SAME BRO
🎂 Thanks for sharing what is coming around the bend! It is always nice to know what to expect in the future. 😉😉❤❤
Polygons - a composition of straight lines. Because the line is a simple (linear) function, that's easier to compute any position on it. NURBs/splines/curves can have multiple control points and weights that affect a position on the line, so it's more demanding on computing a position.
Wow. I can not say anything about the SDF Modelling work as I have been a lighting Artist for many years, This free addon will be a game changer for me. Thank you.
Shoutout o the addon developer for the effort ..! And definitely excited for this feature to be natively available in blender some day.. 🎂🎂
You could cache various "real" versions of that "virtualized" object on first load.
For games that would mean that designers/modellers can focus on the shape itself and the engine can bake/cache multiple versions of this object with vertices on load or even at compile time already.
If it can be optimized enough to work on a larger scale of objects it would basically mean free and far finer control and availability of LODs at any given point in time.
Combine that with lazy streaming of assets and it might not even impact performance that much with a massive gain in more consistent but variable fidelity.
10:34 - Subdiv modeling isn't a lie, it just isn't implemented the way Ed designed it (at Pixar or before?). Real subdivs automatically subdivide to the sub-pixel level, so no matter how far you zoom in, you always get a mathematically perfect view down to the resolving power of your pixels (and yes, SDFs have the same pixel limitation). The fact that devs write faulty algorithms which stop the subdividing prematurely isn't the fault of subdivs, it's the fault of the devs. The same thing would happen if devs implemented faulty SDF code or faulty NURBS/spline code.
🎂Super excited to try out SDF when it's released and I have some free time. I think it's promising. Having tried 3d coat, plasticity/MOI/F360, substance modeler for hard surface concepting - I find nurbs surface modeling to be too restrictive, voxels are great, but remeshing tends to get slow. Sculpting while freeform is difficult to achieve anything non destructive precise, bevels are particularly tricky and time consuming. I do think the best approach is a lossless component editor combined with a lossy polymodeling workflow. In many audio editing DAWs there's a concept of "freezing" that prints to a lossy format. I'd love to see a UI that allows you to freeze/unfreeze the SDF generated components via some sort of LOD or proxy system.
Didn't expect to see a dragon cameo in this one - thanks for the shoutouts!
N-gon supremacy gang !!!😆
also it feels amazing when using triangles on an animated model, feels like you're a professional with the verts
This is a game changer! I switch between parametric, Boolean and polygonal modelling constantly and this appears to be a far more natural approach. The lack of hard edges and conversion to mesh is not as crisp but I'm sure there will be workarounds. I'm excited to see the progress.
Wonderful I was meddling with plasticity and loved it only getting frustrating with not have the same control Im use to with hardops an blender in general. this could be the next evolution. thank you for this awesome info.
I thought your sponsorship explanation video was a cool idea, but seeing it in practice is even better. I’m getting Light Painter today and bookmarking the other add-on
I have been working with vector based software to create 2D stuff, I really wanted to jump to 3D modeling but I couldn't get my head around why I cannot just blend objects and curves like I used to do it with vectors... Ha! so this is why! now I can create with my same mindset but in 3D!!! Thank you!
I think you should also review "plasticity" because it is also a very new and exciting software which is targeted for hard surface and gives you very high level of flexibility and even with very good precision.
This is incredible. My modeling background (hobbyist.. not pro at all) is moslty Solidworks/Fusion 360. Im very accustomed to CAD style modeling. Ive had trouble with poly modeling, but i absolutely see the power of Blender. This could be a great gateway ....
This is awesome, I cannot wait to get my hands on Conjure SDF! Thanks for the information!
I'd think abstract is a better word than virtualized. It's a representation of the mesh before it's turned into a concrete view. Probably stored as some kind of graph data structure under the hood. But that's my best guess.
SDF modeling reminds me of how the game dreams for PS4/ PS5 approaches he modeling process. I think that game / creation enviroment acheves a similar approach with some sort of point cloud modeling process but none the less anything that is as fast and versitile and lets you create actual geometry in the end is very exciting. Thanks for sharing as always
Dreams do use SDFs. You can find some presentations where they explain how they made the game.
Add on is fire!
16:23 light painter looks interesting. The SDF addon seams like it would be fun to make some abstract art with. Interesting video thanks
Might be some misconceptions about procedural materials. They are not in fact, 2d materials wrapped onto a surface, and do not even require UV unwrapping. In fact, if you make a procedural material with some noise to it, and slowly animate the surface shape of an object while keeping the material static with the object's origin, you can see how the material indeed permeates through the object continuously.
The limitation is actually on the rendering engine's side of how it computes the material. If I recall, even Cycles only computes the texture upon intersection of the surface, even if the surface is transparent. I thought there was some improvements coming to this, and volumetrics can provide a solution. I was loathed to ever use UV textures because I felt they were just wrong.
However, for simple things like wear and tear, I eventually learned that a simple rough texture map was far easier and quicker than trying to make a realistic procedural material. I probably could have sped it up with using addons, but I tend to always do things the hard way in an effort to learn how they work, I guess.
🎂 interesting as always. Thanks Curtis
Your component/ modelling pipeline description sounds perfectly reasonable to me.
After some time with Blender I still had to get used to that "virtualised..is it real or not" kinda invisible realm thing you described. It would be good to have a way of formalising that workspace.
The idea of something being virtualized is interesting because it's a concept that moves with the frame of perspective. For example, if the real world is our working environment, then everything on the computer is technically 'virtualized', and the word loses meaning when comparing digital modeling techniques (it doesn't matter whether something is polygonal or not).
But since here we are implicitly considering our working environment to be the 3D view in Blender, using 'virtualized' to compare modeling methods suddenly makes a lot of sense, because it answers whether we have polygons readily-available to manipulate. In the case of a virtualized technique, the answer is usually no. In the case of a non-virtualized technique, the answer is yes.
@@CurtisHolt Interesting. Perhaps, as you say, when a broader range of techniques are in demand in the near future, the necessity of separating composition from casting (to use physical terminology) will be more readily evident.
Makes sense to me. Cool vid
🎂 yay, new video! SDFs looks quite cool! Isn't the new modelling app from Substance SDF as well?
Also, impressive to watch peer reviewed YT video :D
Yep (as far as I know), there are a variety of new (and old) softwares making use of SDF techniques in various capacities. It looks like interest has picked up in recent years.
This is exactly what I am in need of at the moment, when can we get our hands on it?
Seems like a boolean function that joins objects and uses proximity and maybe intersection edges to apply a 3D gaussian blur/average to the vertices. Very impressive and very fast. It would be interesting to be able to choos3 the time of transition/blur, as well as the radius.
Amazing content as always! 🎂
Thanks Spencer!
You're welcome :)
Really interesting.
Reminds me of Real 3D on the Amiga in the ol days.
I agree that it would be ideal for 3d printing, as we seem to be dealing with solid shapes, and there is less chance of missing or dupe poly's. It also looks like it has built in 'cut away' views to examine internal structures.
I’ve been looking into CAD apps, and it looks like this does things in a similar way. Would be great with ‘constraints’ applied.
Be amazing if controlled with Blender animation. Bullet holes strafing a fuselage. Towers growing out of scenery, realtime motion gfx cut reveals of illustrations, bugs running around under the skin.
Spacial textures would be great for cutaways of concrete slabs and building frames.
Interesting to see if the textures would bleed within the fillets applied?
Hey, fellow Real 3D (later Realsoft 3D) user here, also on the Amiga 🙂Thought I was the only one remembering this software.
Thank you so much for sharing info for newcomers to modeling like me! 🎂
There is no cake. 🎂
Great video Curtis, looks really interesting.
then there will be 3d reaction diffusion possible?
Very well described... all of it!
Native SDF would be _so good_ 🤤
Subsurf objects are definitely perfectly smooth mathematically. Not sure if that's what you meant.
holy cow Blender is unstoppable.
🍰 Very interesting. Going to know more about SDF modeling and light painting.
I wonder if noise could be applied to the surface of the primitives?
Very interesting ! Thank you 🥮🎶🎵💃🕺
Hi all, have you tried Clavicula which is free ? This is my personal modeling room and blender is the mesh "operator", my two devices to process.
Thanks for talking about voxel modeling, nurbs modelling, cad modeling, polygonal modeling. And the ultimate SDF modeling.
Which is of course more flexible. Only one thing is missing : it is shader based and the colors are not preserved when meshing is process. Am I wrong ?
Technically metaballs are SDF spheres.
The fact that they are triangulated in blender, just means that internally are not rendered using a raymarching technique, but are rasterized to polygons on the fly to use the same gpu pipeline as meshes.
So it’s kinda like a 3d Version of vector drawing?
Just say this on my recommended and it's wild! Thank you for sharing this!
I have one question though: what's your thoughts on SDF modeling for organic characters? All the examples I saw look like they are for hard surface, but it seems like this method could make molding a character pretty intuitive.
www.youtube.com/@womp3D
So are SDFs editable and can they be used in other programs, video, games?
Yes. Please. When?? how much?? this is a must-have!
its like vector drawing but to 3d objects ?
The analogy of a pixel-based image from a vector image would probably fit.
If in 3D, the polygon based modeling would be pixel-based and then the parametric based modeling NURBS is the vector based.
But these are for the surface or faces of the object.
There is a game called 'world of goo' which uses fuild joined circles to create web type playfeilds. This could be applied here to give some more organic type structures.
Quad remeshers are making big strides. Exo's quad remesh is still great although I don't know when it was last updated. Seems like a perfect use for it.
🍰 SDF is just great. I fell in love with it using Dreams on my playstation. Hope this gets a commercial release.
As someone coming to 3D from a photography background this sounds very similar to the dynamic between PNG & JPG. One gives you infinite detail via mathematical functions that is real handy for clean cut shapes & anything that you want to scale up or zoom in on. While the other is a compromise that is great at capturing the real world quickly and efficiently, but it does this at the expense of loss and general messiness, especially if scaled up or down (pixelation being the equivalent of the subsurf scaling issue you discussed).
Both have their completely separate uses, which is why one format does not dominate and most people develop workflows around knowing when to jump from one to the other - so couldn't agree with you more about your workflow discussion.
Did you mean svg? Png and jpg are both rasterized images that are limited by resolution.
Svg is the infinite detail one.
I would be interested in seeing a demonstration of exactly how one would create these SDF assets
Finally, Ive been thinking of how great it would be to be able to do something like womp3d directly in blender
I need this add-on so so badly 🔥
If I understood right This is why we need to "Smooth Polygon" which means they aren't really smooth but the are being rendered that way 2:53
So it's kinda like using vector lines in 2D art?
Interesting. How does SDF compare to software like Plasticity or Shapr3D?
Component to Model is my workflow when working with Rhinoceros NURBS. After construction, I tessellate to a poly mesh and then export to Blender, Maya, etc.
The bottleneck is cleaning parametric ("naked") edges first before export, since they are mathematical approximations and not explicit as in poly modeling.
so what I'm understanding is it's like vector mapping in Illustrator for example, except in a 3D space. Is that right or am I on copium
hey, how is your condition going man ? I see you moving ur chair around and playing with shapes, i was wondering if that would affect it. Cheers
Oh wow I'm excited that it crossed your mind because I wouldn't have thought that anyone would remember (or care). Yes! Getting used to a swivel chair again has been weird. I've learnt so much about neural adaptivitiy over the last few months.
One thing I didn't expect was that neural expectation for balance is much more context-sensitive than I thought, meaning that you might experience different symptoms simply based on where you are, what you are looking at, even if the exact same motions are affecting your body.
Control is also super important, if someone wobbles my chair, I will get a strange balance disconnect (signal mismatch) almost immediately and feel a vestibular migraine event coming on, but if I make the same motions myself, I am fine. My brain knows exacly which motions are coming, so neural expectation is not mismatched with mechanical signals. Sometimes if I'm eating at a table with people, I have to move away and eat elsewhere because they accidentally create too much movement.
With that in mind, I wanted to test a swivel chair again because it would be more convenient for the space (and I felt like it would be a personal achievement if I could do it).
The first few days were pretty bad, not gonna lie (especially the evenings when I had my lights off, making it harder to keep the room as a frame of reference when looking at the screen). But it seems that by persisting, I was effectively training my brain to learn how much force is equivalent to movement and get used to the control of the chair, specifically in relation to the screen / desk.
Fast forwad to now, I can flop around the place and there are no side effects, however: I feel differently when facing the other tables.
That's the weird thing about the contextual brain training. I feel fine at the main desk moving around on a swivel chair, but stranger on the others, so I can only stay at those for so long.
There's still a lot of work to do (I don't think it's possible for me to ever feel perfect). In this rental, there's a part of the kitchen with an angled corner against a diagonal wall, and when I approach it, sometimes I get weird balance signals and have to back off.
My nan has nice bar stools in her apartment, but they curve down a bit at the front, making my brain get confused and think that the world is slightly tilted forward. So there are still these things I need to be aware of.
You'd think I wouldn't be able to sit or go anywhere but... Cars, trains, planes are fine. It's entirely a contextual tolerance (I still get motion sick in vehicles though, which is a very different feeling to my regular strange signals).
I've also made some good strides in finding dietary things which make me more sensitive to the sensations. Some of it is pretty obvious: alcohol is the worst. It affects the interpretation of the vestibular nerve, so it throws expectations way off-base. Caffeine is also inconvenient but not as bad for that, however after testing, caffeine spikes did trigger new BPPV episodes. They were all anterior canal episodes (weirdly). Not knowing why that happens keeps me testing new things.
Additionally, my resistance to vestibular migraines was good before moving to this rental (after coming off lansoprazole), my theory was that I was absorbing more minerals. However after one of the new anterior episodes, I had an awful vestibular migraine event, it felt like I had lost the resistance to them again. That didn't make sense until I considered that I had been much more active here - walking, sweating in the sun, climbing up and down hills to upload content.
With that in mind, I've been testing electrolyte supplementation (first in isolation, now in parallel to caffine), to see if replacing them makes a difference. It's too early to tell, but I'm hopeful.
@@CurtisHolt it's crazy to me how your knowledge of what surrounds you and your brain interpretation are two separate things, I'm glad you're still pushing, and found ways to make it happen less frequently.
Ya I remember it and cared I've watched the whole video and found the subject quite interesting since I haven't heard about it before !
Great video 🎂
another old 3d program used pointy ended sausage shapes to simulate muscles (like the still suits on David Lynches Dune), which were stacked together like muscle fiber bunches, and then resolved an organic shape when the meta ball like smoothing smooshed them all together.
do you remember how it was called? asking for a research:)
@@ViktorasMakauskas Humm, I posted a link but it may have gone off for verification by the channel.
I think it was metareyes-plug-in for 3dsmax. I used to ponder 'back in the day' that if one defined the 'muscles' that formed the skin with the meta smoothing, that you could then link the 'musles' to the ik and get the 'muscles' to reshape and thus change the skin topography???
@@DaveTimperley heh, some sort of ZSpheres prototype. Thanks!
the only potential problem i see with this video is that you describe SDFs as being "abstract" while polymeshes being "real" - but both are just different ways of expressing surfaces/volumes. This is similar to saying SVGs are "abstract" because there are no actual pixels in the file, but we all know that they are "real" and sometimes even better for some applications.
SDFs are not converted to meshes or point clouds during rendering (unlike SVGs are turned into raster images for rendering) and hence, can be better than poly meshes in some areas - such as CAD / 3d printing / CNC and can have better performance and resolution in raytracing (raymarching).
SDFs are turned into raster images for rendering just like SVGs. All 3 are the same in that regard - polygons, SDFs and SVGs are just descriptions of functions that, at render time, are used to compute the position/color of a pixel. So I would say all of them are just as abstract AND real as each other. Unless you define real as a bitmap.
Great video !
Could you use this technique for character modeling? It seems to work really well for if you're trying to make a machine.
Absolutely, SDF modeling really shines for organic and/or complex forms.
Thank you 🎂
All the native SDF tools need at this point is a straightforward way to subtract volumes, then all bets are off. Geo nodes SDFs are OP and nobody knows bout it. Thx for the vid
I'd imagine you could use discontinuities, areas of high curvature, etc. in the SDF field as a way to place geometry and topology algorithmically, where they'd follow the SDF exactly.
amazing video as usual, but I thing you need more cameras filming you! just one more camera and I can photogrammetry you XD
When will this be added to Blender?
so like modeling with raymarching?
Sphere-tracing styles of raymarching tend to be based on SDFs, yeah.
how do i buy it (SDF add-on), would happily pay like $25-60 on Gumroad even for like alpha access + perpetual key; dont care if it crashes or blows up.
I work in digital avatars/3d printing/organic surf and need something like this right now.
I wonder how it unwraps
Ain't it the similiar method of modelin/3d representation to what povray provided?
SDF was around first with #MadeInDreams #DreamsPS4 since 2013 I personally am making short animations with it! then there are other SDF apps like #Unbound #MadeInUnbound game engine and #MagicaCSG and #Womp3D and #Calvicula and now we are getting it with Blender with #ConjureSDF and some inside Blender tools too! glad to see it's happening for Blender! I personably made some tutorials for Unbound, models with it and short animation with Dreams! funny thing I didn't use it for hard surface at all! I used them for making characters and curved objects! you also can texture paint with it!
i think you should tell the blender guys about that. that really would be an awesome addition for blender 4.0 or something!
No mention of Adobe Medium or Adobe Modeler - pure technical marvel of SDF sculpting and technology.. not to mention the Move and Warp tool which no other app can reproduce - 6Dof move in VR
there is another issue in that its very expensive to continuously calculate all these things in real time. if you could take that final model and then turn it into a standard polygon model that would be good. One problem i had with node based procedural modeling with blender was the fact that it was annoying to convert to real mesh that doesnt melt my gpu.
Doesn't it seem to be kinda similar to svg technology?
This is neat! How muc support for this exists in Blender at the moment? And how good is the conversion from SDFs to poly meshes and vice versa?
Hi, I see the git hub repo for the light paint addon. It contains a bunch of random files and nothing that says 'instal me' what do i need to actually do to get it to work ?
are you talking into your phone or the microphone? I'm a bit confused there
There are also editors for this like Clavicula and Magica SDF
0:15 Oh, Metaballs get improved?
Some time ago I tried to use Metaballs to create effects like these in order to make multiple slime people turn into one.