Video není dostupné.
Omlouváme se.
Crowds in Blender - Ask Blender Bob EP4
Vložit
- čas přidán 25. 07. 2024
- Let's see a few techniques that are used in the VFX world to create CG crowds.
Anima Crowd software:
secure.axyz-design.com
Meshroom photogrammetry:
alicevision.org/#meshroom
I really liked this one.
I never worked with crowds so this was very insightful
This is the best Tutorial I have seen in 2022... Thanks Blender Bob
wow, i am amazed at your knowledge and skill
I think the 1.57 value comes from radian Pi/2 which means 90 degrees
Yes, that is the reason. You could also use a rotate euler node and set it to 90°. That's more intuitive for most people, instead of working with radians
True dat
Those Erindale tuts paying off
Yep, that's it, vector operations are done in radians
Excellent as ever BlenderBob, it's always great to see people troubleshooting and explaining it and potentially being given information from blender community that helps you achieve your goals to pass on to all. love your work :O) hope everything is running smoothly for you, love seeing what you can do with blender rather than just saying nope it can't be done. not going to try mentality! So thanks, really appreciate a true professional putting time and effort into making it work for them! :)
Glad you liked it
You just bring amazing content. Thanks a lot blender Bob!
Awesome and as always with a great sense of humor!
That is the tutorial which I am looking for. Your tutorials has awesome knowledge as always. thanks for share your know how to us
Wow the video geometry was crazy
I didn't know Blender could be used like that. Thanks!
It might seem like i'm overreacting on that but thank you so, so much for taking the time to explain all the vector, euler xyz nodes etc. I've been avoiding geonodes sine the beginning since it always looked so complicated even with respective tutorials (i know i'm missing out on a big blender chunk but heh). Your explanation was so clear and easy to understand, really can't thank you enough for that :]
Very interesting to hear about video-grammetry 🍪
Thanks, as always high quality educational content.
Wow, I didn't know videogrammetry was already used by studios, it's amazing! now imagine in a few years we'll be able to combine that with AI retopo and then even the sky won't be the limit lol
Awesome video as always. Just a heads up regarding rigged people with geometry nodes though. There's no need to export to Alembic. You can use your rigged character meshes. Just put the armatures in one collection and the skinned meshes in another. Then drag the skinned mesh collection into the Geometry Nodes and use that as your mesh instance. You can even use multi-mesh characters (e.g. head, shirt, pants, etc. all as separate meshes) so long as they're in their own sub-collection within the Character collection. For example, if you have a collection called "Characters" you can put sub-collections in that called "Character 1" "Character 2" "Character 3" etc. Each sub-collection will be treated as a unique object. Anima tends to export out single meshes, so this shouldn't be an issue. But if you're using Character Creator to generate mid to close-up quality people, they'll typically come in with all the parts. Also, Geometry Nodes work great for crowds, but particles work just as well, especially for large stadiums. We created a CG version of Bristol Motor Speedway last year and populated it with 150,000 high poly characters using particles as we weren't familiar with Geo Nodes at the time.
Interesting and good to know. But with particles you won’t be able to have the characters to orient themselves to an empty. Unless you know how to do it.
I’m not sure how Blender would manage to handle 120 rigged characters. The other ones are all instances but you still need to update the original ones.
@@BlenderBob It works, but Blender doesn't like it. The company I work for makes theme park rides. One of them is a 120-seat theater called the Motion Theater. We just did a visualization for this ride with 120 high poly (~80,000 poly) rigged people in the seats on the ride. Blender slows to a crawl and is all but unusable, but it still renders. We can improve performance with either geometry nodes or particles, but from what I can tell, Blender still seems to be calculating the armature for each instance. So Geometry Nodes and Particles save RAM on the geometry instances. But the armatures aren't instanced, so things are still really slow. The way we've mitigated this is to save the animation out as either a LightWave MDD or a Max PC2 file and then apply that back to the characters using the Mesh Cache modifier. It's basically the same as exporting to Alembic, but you don't have to export out the characters. Just the motions.
I'm not sure if you've played with Anima's vertex cache export, but you can export your scene using the PC2 vertex cache option. It's slower to set up, as you have to manually load each PC2 file onto its corresponding character in Blender. Once you do that, however, performance jumps way up as Blender isn't having to calculate the armature. It's just reading the point cache off the disk, which is pretty fast. Anima does a nice job of naming the character mesh and the PC2 file, so adding things (while mind numbing) does go by fairly quickly.
For particles, you're correct that you can't orient them to an empty like you can with geometry nodes. For the work we did with Bristol Motor Speedway, we broke the stadium into "zones" (left, right, front, back) and then used separate particle systems to "point" the people in the right direction. The Advanced settings in the Particle controls let you set a phase (rotation) for each particle instance and then a randomized phase, so everyone is pointing basically in the same direction with a little variation. It's definitely not as elegant as geometry nodes and I'd be hard pressed to say that particles are better than geometry nodes. Just noting that you can do it. As with most things, it's just about how you hack it together to get it to work. ;)
Merci beacoup bob vous etez vraiment le meilleur
Videogrammetry is mind-blowing.
so cool great video bob
Awesome tutorial. Very informative.
Now I can do some nice crowd scenes hehe ♥ Thanks Blender Bob
Thank you so much! Amazing information
5:35 "Position" vs. "Location" is the difference between the center point of an object (the object's location) and the precise position of each part of the mesh (or whatever)
I guess both of those could be called "Position" but that difference is used quite consistently throughout Blender and honestly it makes sense
8:32 why are you using a vector scale here when you take a single (float rather than vector) random value and also end up plugging it into a single float value? It's not actually gonna matter much but effectively you are allocating three memory locations for a single value that you also only intend to use as a single value. It'd be more efficient to use a regular math node's multiply for an otherwise identical effect.
13:05 all rotations internally are in radians, not degrees. You can put a degree-to-radians in front or you can type in "pi/2" which is in radians what 90° is. (That's why it's roughly 1.5 - it's closer to 1.5707963.... - I'm guessing 1.57 would basically look spot on. But typing "pi/2" into the field is gonna be exact up to floating point limitations)
Thanks for the tips!
I would guess, not looking at the documentation at all, that position implies also "order" where as location is just "coordinates"
@@matiasguisado3714 "Position" actually refers to a standardized vector in a path tracing rendering pipeline. (Standardized as in this isn't only Blender. It's how it's called in OSL, the Open Shader Language)
You got like three fundamental vectors in a path tracer giving you most of the effects of light bouncing around in a scene:
Position - the ray hit location
Incoming - the direction from which the ray came
and Normal - the surface normal
those three together are really powerful. - So that's what Position really does: If you send a camera ray, it tells you where in the scene, in global coordinates, the ray hit.
Meanwhile, Location is a constant vector: No matter where in space you are, its value is the precise global coordinates of the centroid, pivot point, or origin of an object.
It's for instance very useful to do "Position - Location" which will recenter the global coordinates on the object's center. That's pretty close to what Object Coordinates will give you, except those additionally give you the rotation and scaling of the object.
This channel is flawless !!!
Ah ah! Not if you look at my hair!
really valuable content
Great job! I could never have explained it better than you did at 6:57. 😆
And it's pronounced "Euler", not "Euler"!
Euler not Euler? I thought it was Euler. ;-)
@@BlenderBob Its german, so it is "Ajler" or with french characters in phonetic "Aàilère"
@@BlenderBob oil-er
enfin de retour !
Good stuff man!!! thx
Brutal, Muchas Gracias Master
De nada
Very good video
I love this. Great overview Bob. Have you scene the John Adams Mini Series VFX Breakdown? That was the one I saw years ago that opened my eyes to VFX in general...and crowds.
1.4 roughly half of pi
Blender uses radians for rotation in geo nodes instead of degrees
You can use a math node and set it to Radian and then connect it to X socket of combine XYZ node and then connect it to the Vector math node
Um...sooooooo...CAN we have a Nuke Bob channel as well? I checked...the name is open. 🤣 I would support it in a heartbeat!
Did you check my compositing series?
@@BlenderBob I sure did, sir! You're the one who finally made color space click in my head. Soon after watching your series I accepted my first freelance project that had an ACES + Linear workflow...and I didn't do too bad. But that series played a big part in my first project under a Nuke Indie license! 🙌 So thanks for that!
But anyways...I'm just greedy and want more nuke LOL. 🤣
please please please make an in depth video about Videogrammetry and how to use it. It looks so awesome and I would like to know how it works!
End of December or early January
top Bob.....
When working with those characters and importing them to blender I think It would be far better to use another blender file.
- You do your thing in Anima Crowd software and export FBX files.
- You import them into a blender file called "crowd assets" where you put them in a collection and clean up naming, materials,....
- And then you link the objects or collections from "crowd assets" file to the Blender file in which you actually have your scene
Well, the actual pipeline is actually more complicated than that. This is a 15 min overview. But thanks for the advice. :-)
wow. thx
"How it could be done in blender..." yeah, i'm most interested in is how to get a job with it :D
Would like to know more about the vectorpass, cryptomatte(1pixel of boundary is coming) and depth pass(not as perfect in native in comp) for defocus and shadow pass and AO(empty space has black unlike other apps) pass
👍👏
great video
if you know where to find models equivalent to anima 4d compatible with blender, I'm interested
Hey! :)
Did you try creating a mask using a color vaules (like a keying) from diffuse texture to make different roughness/specular? :)
(Talking about the videogrammetry)
Cheers!
Yeah, we did. We actually had the characters dressed with flat pure colors so we could make some variations.
3:50 could merge by distance ✨
That’s what I do at 3:36
I don't know where you've been hiding but we're friends now.
I’ve been here for a while now. Question is, where were you?
@@BlenderBob Dude, I was here. What were you wearing? I'm in jeans and a jean jacket.
Amazing...just on time..I've just used anima and it was a struggle...Thank you.
Also, what do you think about Human generators for mesh, and a real-life captured animation?
Never tried it.
Hey Bob! Super cool video! Any reason why you can't just move the meshes only into a new collection after the FBX import? Rather than moving everything (including Armatures) into the collection, if you just had a collection of the meshes, the geometry nodes should still be able to select a random character mesh, or is this wrong?
Actually, you can but only if your character is in one mesh. If not, you need to make a collection into a collection. The advantage of using alembic is that you don't need to calculate bone deformation for 120 characters. No way Blender can handle this. It's also better if you need to send to another software, like Houdini
@@BlenderBob great info! Thanks Bob!!
I discovered this amazing channel today, why not before? Because it's EPIC!!! The Pixel F#ck video is my all time favorite!!!
Any thoughts on randomly offsetting the animation of the character per distributed copy? The only way I've discovered to do is to do it manually by cloning the source and offsetting the animation which would naturally bloat the scene file. If I've got a 200 frame sequence and a 1000 frame animation of the same character, creating such a large number of offsets by hand to set this would be a huge job and seriously increase the file size. Something like even Superspray in max 2.5's particle system could do this so I'm still unsure as to why we don't have this in B3D yet... unless I've missed something... thoughts?
You can’t. And I spoke to the devs and nop, they confirmed that it’s not possible. Probably when they make an alembic geo node reader. So our solution was to user Clarisse as a renderer (the demo here is all Blender) and even there, the Clarisse guys had to make a special custom version for us because getting a random offset was already built in but you also need to offset the texture sequences and that was the tricky part.
Hi @BlenderBob At 15:14 you talk about an animation that changes the mesh for each frame. How do you do that? Do you have a different block of data and switch between them? do you use a plugin? I'm very interested in this technique to apply it to some clay animation tools we're working on. Thanks in advance!
It’s part of the videogrammetry process. Each frame is different and they are exported as alembic.
7:47 you don't need a vector math node to scale down random values if you going to convert in back to float anyway. Just a math-multiply node will be is enough.
I used to do that but then you need to adjust 3 values, unless you create a value node.
@@BlenderBob Where did you get 3 values to adjust? You are creating 2 random float from -1 to 1, scaling them down (multiplying) it down and combining to xyz(xy) vector data. Maybe you intended to use random vector data, scale it and compress to single float number for both x and y?
Aesome @blenderbob great great one!
With the videogammetry part, having the topology change every frame and the need to have new textures and uv:s for every frame seems a bit rough, couldn't you for instance use wrap4d on the face to project the changing geometry into a base mesh (the animation will be transfered to blendshapes) and then use for instance Houdini's point deform to deform a static geometry to the movement of the videogammetry model as it works based on distance rather then point numbers.
This is the workflow I have used when having to deal with changing topology and this way you can have different parts with different materials and not needing to have a thousand textures for 1k frames of animation.
We tried but it doesn’t work. For example, if you have two people fighting each other, they touch so it’s not possible to separate them. Or if someone has his arms crossed then they are merged with the body. Or there’s too much changes in the cloth or hair topology. Look at the guy dancing in gray in my clip.
I spent weeks in R&D on this. I’m resuming it in 15 mins. We had so many issues that we had to resolve, some of them directly with the developers
@@BlenderBob Some of those are issues of the tech yes but others are solvable. For the crossed arms example, it really wouldn't be an issue as long as you separate the arms on the mesh that you apply the point transformations to from the videogammetry.
When it comes to hair and cloth, I simply don't see the benefit of having them be captured in the videogammetry, that just makes them unadjustable for really no benefit whatsoever when you can just run a basic cloth sim & hair sim after the fact.
At least for photogrammetry in general, hair will never look good as individual hair strands are way too thin to be captured by any camera in a way that they could be calculated using photogrammetry, and like with cloth there really is no need when you can wear a hairnet and just add the hair later.
I'm also curious how you are dealing with the fact that with videogammetry with changing topology, you are fully locked in to that exact framerate as you really cant interpolate the animation at all without complex setups so that it doesn't break the textures.
It just seems like including those things add an unnecessary burden of limitations when compared to the alternative of just capturing performances as videogammetry and simulating the rest.
Id love to chat more about this as I think that videogammetry can truly make for some spectacular looking hero assets.
Edit. To add on the people fighting thing, also really don't see that as an issue, just separate the actors in the initial static frame and then set the point deformation distance to a value where the base mesh doesn't pick up the animation of the other actor when they get close, you can even adjust it frame by frame.
Well, it sounds simple but it’s far from being that easy. You are capturing a performance. You need the facial expression. You need the cloth deformation. We tried remapping on another geometry and we failed. We spent months on this. And vidéogrammetry is not designed to be seen in closeup as a hero model. Too many artefacts. Try to find an example online. Some companies have demo files. Try it and you will see
@@BlenderBob Did you try Wrap4d? its specifically designed to remap facial volumetric sequence captures to a single mesh.
And videogammetry works for hero assets if you treat it as kind of a glorified motion capture where you extract the motions from the captured data itself rather then use the actual changing topology in the final render.
This is to the benefit that you obviously get a hero level asset from the photogrammetry that matches 1:1 with the performance.
I work a lot with Houdini and especially fluid sims so dealing with changing topology is really common for me, it just needs a pretty tight pipeline but getting something usable out of a volumetric capture definitely is not impossible.
May i ask what software did you use to try to remap them into another mesh?
I actually talked out of my ass for some reason when it comes to Houdini, I'm so used to the process so i totally forgot to mention that you obviously need to create a reference frame from the mesh that has changing topology that acts as a rest state for the base mesh that you are transferring the motion to.
But transferring changing topology to a single mesh is all and all really not that complicated.
If you have a small sample you could share, i would love to take a look and see if there is a way to make it work for you.
But in all honestly, it would be a lot easier to just use the camera setup to first do a scan of the actor and then use it for a markeless motion capture as you already have so many angles to basically make the motion capture super accurate.
Edit. Also to add, adding motion blur to changing topology is super easy, just store the point velocities inside a three vertex maps.
Thats atleast the suggested way, though dont know if cycles supports getting motion blur values from vertex maps as I mainly use Octane and Redshift.
Edit 2. If you transfer the animation from the videogammetry to a single mesh you can also then rig the single mesh and transfer that animation to bones of the rig, bake it to keyframes and now you can even adjust the animation.
Great lesson, thanks!
Tell me how to set the animation time interval for clones? I can't seem to find a way....
Actually not possible. They are not going to fix it because the entire particle système will be rebuilt from scratch to work with geo nodes, Then hopefully it will. So you make many versions of your animation.
Well... will wait for updates, thanks for the reply!
love your tutorials so thank you!! cards in VFX are same as Blenders images as planes right?
Yeah, just a single polygon with a texture on it.
@@BlenderBob Thanks for replying Bob :)
pls make nuke tutorials sir u r the best love from india 💌💌💌
I did some already but I’m just mid level in Nuke
How do you make good looking fire / explosions in blender (not 3th party software). I know how to make fire and smoke but never get it convincing are there some rules or so one could use?
I’m not a specialist for that kind of stuff. At the office we just use Houdini because nothing beats it for simulations
@@BlenderBob ok it's what I see at othe yourubes despite there quite an advanced solver now. I think a few explosion sizes / rockets/fire would be fun. But it's okay I'll keep trying
14:56 I like you Blender Bob you are nice.
Hello Bob! Love your tutorial. Need something simpler and quicker for now. Spawning planes along a geo via particles. Using a library of 2d crowds i already have. Do you know of a way to randomize the start frames on the crowd sequences? Complete Blender newbie and impressed how quick i got there. Now its just they all start on the same frame sadly….
No idea. And now it’s the first beautiful day of the year where we can wear shorts and I don’t feel like being in front of my computer. ;-)
Hope you enjoyed the weather! Me and my son‘s hike to a new waterfall today was marvelous. In case you stumble across a solution for my issue, let me know! I have gotten zero help/feedback in any of the groups i found and fear i am just not experienced enough to try your 3d approach. 😅
@@MrCosmopolite Yeah, you can't control the image sequence per object, that I know of. We found a way to do it for the alembic when we use videogrammetry but it's quite complex and it uses geo nodes. It wouldn't work in this case.
What is the videogrammetry setup like? Is it crazy expensive?
It’s insane and quite expensive. You can find some setups if you google it.
Guess would save a lot of time if anima could export into alembic. Wondering why they have not implemented that already. The software has been out there for years
Anima will export to PC2, which loads into Blender just fine. When you export your scene in Anima, you choose Blender as the file format for the FBX. Then import that FBX into Blender. Add a Mesh Cache modifier and then copy that modifier to all the characters. From there, you load in the PC2 onto each corresponding character. The mesh character's name will match the PC2 file's name, so it's pretty straightforward. I just added 100+ characters to a scene in Blender and it took about an hour to get the PC2 files added. It's still slow, though, because Blender isn't optimized for this sort of work. You can bring the same number of characters into Unreal with the actual skeleton and it works flawlessly.
@@SteveWarner not sure why anima doesn't then do alembic export. That would so much faster
Another great tutorial. Btw, is there any chance to know the "secret partner" in future, or they will keep "secret" 👽
Not a chance! :-)
@@BlenderBob 😢
you don't need to bake them to Alembic, put each character in a collection in other file. link that collections in current scene and just use each of linked objects as a single object for distribute.
Yeah but Blender can’t handle 120 rigged characters at the same time.
@@BlenderBob For complex rigs yeah. but for simple rigs like you used in tutorial will be faster than Alembic. and benefit of using of "linked collection" is you can hide rigged mesh in view port and use a proxy mesh that is hide in render.
I guess you used some python to set all those models in to bounding box or was that alt click?
That alt click won't work for me as i use a Wacom tablet and need alt to pan, scroll and do other functions. Guess I need to make a custom toggle for that option so I can use it on the fly
No. You just do it in the original characters and all the instances will change too.
How do you make blender show proxies instead of actual geometry? Thanks for the video
Go in object properties, viewport display, display as box
@@BlenderBobThanks.
Is it possible to apply displacement effect on crowd
Huh?
How much would you charge to for a scene like the stadium for a music video if I do green screen at home
Can’t u just use that one you made and put me in it
@@kidkreation I can't because it's only ten different people, I can't show them from the front. I cost us thousands of dollars to make that clip. In order to do this for a client, we would need to redo the videogrammetry for all characters, like 75 at least, and it's 500$ each. Plus 3D setup, lighting, rendering layout and all, about 60k.
I once had to fill a gladiatorial arena with a crowd from only 6 filmed groups of around 20 extras. Using cards (We just called them patches). Took about a month and I still felt I could identify duplicate actors and motions. We had to cheat the camera focus to hide the problems and the grade was insane. So difficult to do the old way. Great tutorial right here!
(The vector add value of 1.5 is certainly unusual. Does anyone know why this happens? I have come across similar issues and while the solutions do work - I always like to know why something works and I cant work this one out!)
1.5 comes from using radian instead of degrees
@@BlenderBob Sounds reasonable!
@@ianmcglasham Yes, as Bob already said, Blender works with radians internally, so 90° equals Pi/2 which is roughly 1.57
Please Please Please make a separate Video on Videogtammetry....
In January most likely.
@@BlenderBob Thanks in Advance.🙏
pls tell me the scope of blender and motion graphics in CANADA
Motion graphics, no idea. Blender, strong before Tangent's bankruptcy. Now, besides Real bt Fake, I don't know any other companies that are using it, unfortunately
Feel free to be nuke bob. I want to see it. I don't have nuke, but I'm so fascinated by compositing.
agree
Have you seen my compositing series?
Wait. What's your system configuration here to be able to do that at home?
My system at home is irrelevant as I’m only using it to display the office’s screen on my machine. At the office it’s an AMD ryzen 32 cores 64 gig ram and 2080ti. At home iMac Pro
@@BlenderBob When someone says 64 Gigs of RAM .... my head hurts.
@@nirmansarkar When I started in 95 I had 64 megs of RAM and people thought I was talking about my hard drive
I'm sad when the anima guy sad
15:50 What is CPU RAM? Is that just your computer's regular RAM?
Yes
@@BlenderBob thank you!
Hey Bob, please try to recreate the head explosion vfx from the movie ”the boys"
Got a link?
czcams.com/video/N0LWKcYyWl4/video.html
2:54 wait is blender bob a french speaker?! C’est intéressant ça
C’est ma langue maternelle
@@BlenderBob moi aussi, mais t'inquiète pas c'est pas la seule chose que j'ai trouvé intéressant 😂
I like the idea of videogrammetry but the problem is that there is no free way to do this and websites like renderpeople are way too expensive i could not find a way to get the people to recreate the concert shot you did
Excellent video, thanks Blender Bob! I My girlfriend and I really look forward to your videos, you're our fave! We always learn a ton, and your teaching style is accessible and entertaining! We for one would love to see a Blender Bob take on some comet-hits-earth planetary destruction, would love to learn how you might approach this! I loved how they did it on "Don't Look Up" - czcams.com/video/4-zv5Cvg6pM/video.html. Missing from this clip are the floating chunks of Earth with foliage on top and rock/dirt below, which to me sold it the hardest, but I couldn't find a good clip. Thanks again Blender B, looking forward to the next one, hope life is treating you great!
That’s exactly what we are working on right now but it’s a mix of Houdini and Blender.
@@BlenderBob Rad! I for one dabble in Houdini and would love to learn the workflow from you, but I know you are not Houdini Bob. 🤖🤖
Can u make the foundation’s title opening with all those particles and sand color thing or halo opening title with the sand turn into the suit , please please
Link please
@@BlenderBob czcams.com/video/iWgZBFdRgCo/video.html
czcams.com/video/o6qm88h9XQA/video.html
@@BlenderBob and thanks for the fast reply, I think u r the only one on CZcams that really considers our suggestion
Do you have a link about the FX you want me to show you?
If possible make a video on render passes.
It’s on the list
I just need to get some time to do it
Ian hubert had 1 minute video doing same with different technique
Blender geometry nodes must make the hudini guy happy he's not getting as much simple stuff thrown at him. So he can focus on the more complicated things.
Also I think there might be a setting to tell blender to work y up but I may be mistaken
Ah ah. No way you can make Blender Y up
Make Nuke Bob!
Nop! Sorry! I did make a few clips on compositing though
@@BlenderBob
Thanks for the response. I wasn't expecting you to read it. I've watched a couple of videos, I'm a complete amateur novice, just tinkering with Resolve Fusion. But I came to watch specifically to see the magic trick of the composition at the beginning, but the node tree is intimidating. Your videos are very interesting, and I like your style. Greetings from Argentina.