this is really awesome!!! Wondering how long these simulations take to calculate, and is it possible to achieve real time fluids with this method if the particle count is lowered?
@HarviuLee9 ай бұрын
I was reading their paper, the beach scene is about 3fps and the fountain is about 10fps. But their technique is on CPU, I think the current GPU can handle real-time fluid simulation depending on the number of particles.
@coolkattcoder9 ай бұрын
@@HarviuLee Wow! With all the recent and ancient fluid physics optimizations, I wonder how fast it would be possible to make fluid physics if all the optimizations were combined? It would probably be nightmare to code though lol
@petterlarsson7257 Жыл бұрын
you just said instable what
@lalalu4857 Жыл бұрын
Amazing！how did you complete it？
@Fireheart318 Жыл бұрын
Hopefully we'll be able to do this kind of stuff in real-time one day!
@spoiledoats2 жыл бұрын
@Superku2 жыл бұрын
😯
@x_chow1692 жыл бұрын
3:48 best part
@x_chow1692 жыл бұрын
2:43 iya my favorite part
@3danimationstudio7332 жыл бұрын
kzread.info/dash/bejne/eqttrKZqobnHebQ.html
@BruinChang2 жыл бұрын
What is divergence-free?
@NocheTenebri2 жыл бұрын
This doesn't have anything to do with the video, but what is that dragon model? I ask because I own a resin dragon figure that's identical to it and would love to know more.
@PunmasterSTP2 жыл бұрын
Divergence-free? More like “this has got to be”…one of the coolest videos I’ve seen. Thank you for simulating and posting!
@xayah33752 жыл бұрын
Breathtaking
@inhtien52812 жыл бұрын
Can you give me the link to download the software?thanks you [email protected]
@inhtien52812 жыл бұрын
Can you give me the link to download the software?thanks you [email protected]
@ExplosiveAnanas3 жыл бұрын
Splash! And the bitrate went bananas
@weiren87463 жыл бұрын
Fantastic!!!
@MinogFarted3 жыл бұрын
When the monke is sus!1!1!1!!1 😳😳😳
@slitthroat62092 жыл бұрын
Where monke
@Volt64bolt3 жыл бұрын
Can I do this in blender?
@uyt43 жыл бұрын
fie comment
@butterballshorts29943 жыл бұрын
Here since everyone is gonna get recommended this
@awanreceh12943 жыл бұрын
Why there's only three comment here
@z6lda3 жыл бұрын
cool m8
@HotHello773 жыл бұрын
beautiful work!
@WhiteDragon1033 жыл бұрын
7:19 The way I normally deal with oscillations in constraints like these is to subtract from the force the relative colinear momentum between the two particles. This intuitively means you're trying to constrain the velocity, rather than the position, such that at T+1 the constraint is satisfied. If two particles are already at the correct distance, but there is some velocity that would separate them at T+1, the constraint would remove those differences. (You'd be limited to constraining co-linear velocities; that is, the component of velocity at time T that is parallel with the direction between the two particles at time T. Otherwise you'll introduce torque inadvertently and violate conservation of momentum.) The approach you're taking is that the position of a particle is influenced whereas its velocity is not. I think this is physically incorrect. I suspect it could lead to energy loss.
@tassilo1443 жыл бұрын
Hi Zergling, thanks for you interest in our work. Maybe my intuituve explanation with the analogy to the distance constraint was a bit too hand waving. I will try to clarify things by pointing out explicitly how this analogy is meant: The distance constraint is a 1d constraint (has a scalar constraint function) and acts on two particles. The density constraint is similar but more complicated. In the continuum we have one constraint equation for each position in the fluid and the constraint acts on all velocies of the fluid. This means the constraint is "infinite dimensional". In the computational setting this is approximated on a MAC grid. There it is a k (number of active pressure dofs on the grid) dimensional constraint that acts on the velocities of all active grid cells. Similar to solving the distance constraint where the position or velocity updates can be found easily, the density constraint can be enforced by solving the pressure Poisson equation (PPE) on the grid. By applying correction velocites the distance constaint can be solved exactly (within the floating point tolerances). For the density constraint we have two problems. First, solving the linear system from the PPE to a very high accuracy requires a lot of computation time. And second, even if we do so, we still have the discretization errors from the grid. They can only be reduced by using finer grids or more advanced interpolation methods which also results in high computation costs. So if we want to have fast simulations we have to live with an approximation. Your suggenstion to update the velocities such that the constraint is satisfied is the correct solution and is exactly what we did in the single PPE case that shows the explosion artifacts. Regarding your second question on energy loss: Correcting the particle positions does not change the velocities and the kinetic energy does not change. Since we are simulating incompressible fluids there is no such thing as a compression energy that changes. Only the gravity potential changes. In most cases it should add energy but in my intuiution it should not be able to add much more energy than we lost due to the density errors. In our understanding the compression is purely an artifact because of discretization errors which is unphysical in the first place. Not adding the velocities of the density solve is not wrong because this velocity field only consists of divergence components. Advecting the fluid with these velocites will result in a constant density field in the next time step. If there would be no numerical errors in the pressure solver all this added velocity would be removed by the pressure solver in the next time step anyways because then the density error is 0 and it enforces a divergence free velocity field. So not adding it and directly moving the particles just helps the pressure solver to remove it again. Of course this solution is not perfect but it works well in practice. I hope this helps. Thanks again for the questions.
@WhiteDragon1033 жыл бұрын
@@tassilo144 Thanks for the detailed response! I admit that I don't have much experience with fluid simulations. I am coming from the perspective of working with distance constraints, where constraining the position such that there is no change in momentum would potentially result in a net change in angular momentum. I think it'd have an effect similar to angular drag. I'll be honest, this conversation is making me want to take a crack at writing my own fluid sim, haha. Maybe it'd help me understand your solution better. The most I've done on that front in the past was giving each particle a radius, and then forcing any two particles apart if they were overlapping using distance constraints. If the constraints were weak enough they'd kinda sorta act like fluid, but really compressible, even if you used extremely small timesteps.
@extrauser14984 жыл бұрын
0:10 0:27 that's not even a broken dam breach bruhhh
@PersianEmp1re4 жыл бұрын
Amazing
@chaoticprogramming4 жыл бұрын
0:43 What happend there?
@abbynievs24884 жыл бұрын
Me: she My daddy: he
@MisterTyken4 жыл бұрын
What brought this on?
@YYYValentine4 жыл бұрын
Nice! Is it only for physics simulation, or rendering too?
@6rnh3424 жыл бұрын
poggers
@dragonfly02444 жыл бұрын
No one: IISPH: *S E I Z U R E*
@yank2002ify4 жыл бұрын
Waiting for “Why this was recommended to me?” guys. I’m with you all
@g.s.r.ribeiro69044 жыл бұрын
2016
@stonefreesia69304 жыл бұрын
I can't stop watching these please help
@richardblake35203 жыл бұрын
I think u satisfied to watch this
@momasosenvideos78954 жыл бұрын
poor gpu
@cornagojar4 жыл бұрын
well done
@greedfox78424 жыл бұрын
nerd. >:3
@vccharan4 жыл бұрын
How do you record the simulations on the splish splash software?
Пікірлер
How canyon scene is made? Any tutorials plz
Nice!
How do you make the 3D scene
this is really awesome!!! Wondering how long these simulations take to calculate, and is it possible to achieve real time fluids with this method if the particle count is lowered?
I was reading their paper, the beach scene is about 3fps and the fountain is about 10fps. But their technique is on CPU, I think the current GPU can handle real-time fluid simulation depending on the number of particles.
@@HarviuLee Wow! With all the recent and ancient fluid physics optimizations, I wonder how fast it would be possible to make fluid physics if all the optimizations were combined? It would probably be nightmare to code though lol
you just said instable what
Amazing！how did you complete it？
Hopefully we'll be able to do this kind of stuff in real-time one day!
😯
3:48 best part
2:43 iya my favorite part
kzread.info/dash/bejne/eqttrKZqobnHebQ.html
What is divergence-free?
This doesn't have anything to do with the video, but what is that dragon model? I ask because I own a resin dragon figure that's identical to it and would love to know more.
Divergence-free? More like “this has got to be”…one of the coolest videos I’ve seen. Thank you for simulating and posting!
Breathtaking
Can you give me the link to download the software?thanks you [email protected]
Can you give me the link to download the software?thanks you [email protected]
Splash! And the bitrate went bananas
Fantastic!!!
When the monke is sus!1!1!1!!1 😳😳😳
Where monke
Can I do this in blender?
fie comment
Here since everyone is gonna get recommended this
Why there's only three comment here
cool m8
beautiful work!
7:19 The way I normally deal with oscillations in constraints like these is to subtract from the force the relative colinear momentum between the two particles. This intuitively means you're trying to constrain the velocity, rather than the position, such that at T+1 the constraint is satisfied. If two particles are already at the correct distance, but there is some velocity that would separate them at T+1, the constraint would remove those differences. (You'd be limited to constraining co-linear velocities; that is, the component of velocity at time T that is parallel with the direction between the two particles at time T. Otherwise you'll introduce torque inadvertently and violate conservation of momentum.) The approach you're taking is that the position of a particle is influenced whereas its velocity is not. I think this is physically incorrect. I suspect it could lead to energy loss.
Hi Zergling, thanks for you interest in our work. Maybe my intuituve explanation with the analogy to the distance constraint was a bit too hand waving. I will try to clarify things by pointing out explicitly how this analogy is meant: The distance constraint is a 1d constraint (has a scalar constraint function) and acts on two particles. The density constraint is similar but more complicated. In the continuum we have one constraint equation for each position in the fluid and the constraint acts on all velocies of the fluid. This means the constraint is "infinite dimensional". In the computational setting this is approximated on a MAC grid. There it is a k (number of active pressure dofs on the grid) dimensional constraint that acts on the velocities of all active grid cells. Similar to solving the distance constraint where the position or velocity updates can be found easily, the density constraint can be enforced by solving the pressure Poisson equation (PPE) on the grid. By applying correction velocites the distance constaint can be solved exactly (within the floating point tolerances). For the density constraint we have two problems. First, solving the linear system from the PPE to a very high accuracy requires a lot of computation time. And second, even if we do so, we still have the discretization errors from the grid. They can only be reduced by using finer grids or more advanced interpolation methods which also results in high computation costs. So if we want to have fast simulations we have to live with an approximation. Your suggenstion to update the velocities such that the constraint is satisfied is the correct solution and is exactly what we did in the single PPE case that shows the explosion artifacts. Regarding your second question on energy loss: Correcting the particle positions does not change the velocities and the kinetic energy does not change. Since we are simulating incompressible fluids there is no such thing as a compression energy that changes. Only the gravity potential changes. In most cases it should add energy but in my intuiution it should not be able to add much more energy than we lost due to the density errors. In our understanding the compression is purely an artifact because of discretization errors which is unphysical in the first place. Not adding the velocities of the density solve is not wrong because this velocity field only consists of divergence components. Advecting the fluid with these velocites will result in a constant density field in the next time step. If there would be no numerical errors in the pressure solver all this added velocity would be removed by the pressure solver in the next time step anyways because then the density error is 0 and it enforces a divergence free velocity field. So not adding it and directly moving the particles just helps the pressure solver to remove it again. Of course this solution is not perfect but it works well in practice. I hope this helps. Thanks again for the questions.
@@tassilo144 Thanks for the detailed response! I admit that I don't have much experience with fluid simulations. I am coming from the perspective of working with distance constraints, where constraining the position such that there is no change in momentum would potentially result in a net change in angular momentum. I think it'd have an effect similar to angular drag. I'll be honest, this conversation is making me want to take a crack at writing my own fluid sim, haha. Maybe it'd help me understand your solution better. The most I've done on that front in the past was giving each particle a radius, and then forcing any two particles apart if they were overlapping using distance constraints. If the constraints were weak enough they'd kinda sorta act like fluid, but really compressible, even if you used extremely small timesteps.
0:10 0:27 that's not even a broken dam breach bruhhh
Amazing
0:43 What happend there?
Me: she My daddy: he
What brought this on?
Nice! Is it only for physics simulation, or rendering too?
poggers
No one: IISPH: *S E I Z U R E*
Waiting for “Why this was recommended to me?” guys. I’m with you all
2016
I can't stop watching these please help
I think u satisfied to watch this
poor gpu
well done
nerd. >:3
How do you record the simulations on the splish splash software?
*cough* ...maybe implement it in Blender? *cough*
2:08
no!!! stop hurting the rabbits!!!
Nasa called. They want their super computer back
it only needs a single pc
You need to add an random movement to the water
Pause the video and look at the dragon in 3:41
Derpy sneaky dragon
the water doesn't stick to the container
Because its water