How Do VFX Artists Make Battle Scenes With Giant Crowds? 

yep the guy is super enthusiastic, love it!

Same. I always watch VFX Artists React from Corridor Crew and I love that so I’m so glad I found this channel. It’s literally movie magic to me.

same here! just subscribed

Interesting perspective - I mostly agree with this although there are a couple things I disagree with too. You're 100% right in that a lot of the time what makes VFX stick out like a saw thumb is sub-par animation, especially in regards to humanoid rigs (because obviously we are human and spend most of our time observing other people so its easier to notice when it looks awkward/off) so for realistic actions accurate and clean motion capture data is optimal and can save a lot of time. You mentioned Left 4 Dead so I'm going to comment on that example first, for L4D1 Valve used and combined hundreds of motion capture takes for the many actions of the zombies, locomotion, hit reacts, death animations etc however the compromise of this is that a lot of it isn't cleaned up to the highest quality this is especially noticeable when you focus on things such as the fingers for example which stay in the same static idle pose and as far as I can tell aren't even animated - so yes Valve achieved much more believable animations but I don't think its exclusively because they had "higher quality" motion capture to use but instead they had a much larger variety of data to go off and use, also things can get a bit finicky when using video games as an example because there is a whole other technical process involved in actually implementing these animations, anim graphs, state machines, locomotion frameworks such as advanced locomotion system (which a lot of AAA games use now), etc. Moving on to your Avatar example one thing that actually sticks out in the first Avatar movie is that in some of the scenes when the Na'vi are moving at high speed it can actually look a bit awkward especially when compared to the most recent installation in the franchise - this is because the Na'vi obviously have much longer limbs than the actors in motion capture suits, so when for example an actor sways their arms really fast it can look oddly quick because the length from the origin of the pivot (the shoulder/upper arm) all the way down the arm to the hand is much longer - there is a particular scene when Jake (in his avatar form) jumps from the top of a water fall into the water down below which stands out to me the most. Also avatar isn't all motion captured - things such as creatures especially were hand keyed by animators at ILM and sometimes you don't get a good result from the mocap data for whatever reason so hand keyed animation is still involved. And that kind of brings me to my main point, I feel like you downplayed the importance of hand keyed animation in the VFX industry but productions such as Chappie, District 9, Jurassic world, The Mandalorian (the season 1 speeder bike chase comes to mind specifically) and of course the Transformers franchise all feature a large amount of hand key animation and visual effects studios such as WETA digital and ILM often require you to have experience with both motion capture cleanup and hand keyed animation because sometimes motion capture just isn't always an option such as if you're animating a physically impossible action or something such as a dragon for example. So motion capture isn't absolutely necessary - what is necessary is good animation and a team of talented artists can often achieve that.

​@@Megasus-ju1mg thanks for the response regarding L4D, I didnt expect they would have an animation for fingers, i was focused on general movement, and imo, that huge library helps *a lot*, same as with LoTR. They tried to implement so many variations that it just looks really alive. And you are right, you need to do plenty of cleanup and fixes. But I think this is where the slippery slope lies? You need to do a cleanup, but it is possible to make it too clean to the point that movement are nice, but slightly odd or too smooth, or how to explain it. Or you clean some keys and then the interpolation will make it less real. Maybe you want clean it so there is a bit less data and jitter, but you can also loose some of that life in it. But I think that might not be the main thing. If you look at modern crowds, most of the time, when they walk, they have this rigid NPC style walk that you see in games, but not in reality and it looks more like they overlayed general animation to whatever character. Generic walk cycle captured by man is applied to women, walk cycle with arms wide apart so the animation fits to any kind of person, thin, thick, plenty of clothes, lack of clothes. I didnt mean to downplay hand animation. I meant to talk only about Na'Vi/Avatar mo-caps/performance-caps. Maybe I should have pointed that out. Because yeah, rest of stuff is hand animated because there probably is not a way to motion capture various kinds of fish (i guess one could, but.. why would anyone bother, if you can do a good hand animation quicker and believable) or birds or creatures, not to mention six legs or limbs or dozens of tentacles.. or infamous Smaug performance session that ditched the data and just hand animated the dragon. And pretty much any monster in the movie is hand animated. And I'd guess that most background apes in Ape trilogy is done just by hand.. and especially fight scenes are surely pure hand animation (also even if you have some reference, you still need to hand animate it). It is a neccessity and can be done so good that you dont even question it. (and there is also rotomation, which is pretty much matching actor frame by frame with 3D character). With mo-cap you are limited to reality, with hand animation, sky's the limit. Which brings me back to Avatar/LoTR vs many other movies and focus on human(oid) characters, beause that's when we notice it the most (and also who knows how many shots did have hand animated character in Avatar, cause something either didnt work or needed a big fix). I guess the main difference is, in the end, between tailor making your mo-cap to what you are actually needing, vs having general library of general movement for general crowd, so you can pull from that library and apply it to any crowd simulation you need. Elves? Put there the general mo-caps of stoic people. Orcs? Put there the general mo-cap of brutes. Walk cycle? Put there the general mo-caps of people walking. Instead of tailor making every mo-cap to the precise nature of your need. I mean, it makes sense, it saves time, money, hours, and it is a thing that most people dont notice or care about. But I think that is that one last percent difference that grounds it even more.

@@Patrix8558 Yee that makes sense, and yep you're definitely correct in a lot of scenarios motion capture is more efficient because as you mentioned its quicker and also you are "constrained" to reality since you know you have accurate data from someone physically doing whatever actions are needed in person. and your point about cleanup also has a lot of merit, at Ubisoft for example they don't actually delete any of the underlining keys from mocap date, they just mould tweaks around it by tweaking curves and using animation layers in a none-destructive fashion and I'm assuming thats how most people do it too - someone inexperienced with mocap can definitely have the tendency to "over clean" it when for realistic motions you need those imperfect arcs and slight micro movements. Just felt the need to underline the importance of hand-keyed animation in the pipeline as, in general, I feel like a lot of people (especially those outside the industry) often make the assumption that minimal work is required when it comes to realistic animation in VFX because "the computer does all the work" - not saying that's your interpretation but I just wanted to highlight that.

Honestly, I highly doubt the motion capture tech of over 20 years ago is that good by modern standards. You can clearly see it in the video when they show the animations they used. They are a bit floaty and not everything is cleaned up properly (it's just that it does not necessarily matter for tiny orcs in the back of the shot). The systems themselfs decent but nothing to write home about in this day and age. I think it is more the variety and the quality of pieces themselves that made those mocaps age well. Vicon launched a 1 megapixel camera in 2000, a 4 MP camera in 2004 and a 16 MP camera in 2008. I literally was in a budget mocap studio that still used old 0.3 megapixel cameras as late as in 2016. I am sure Weta could afford those premium cameras but today 1 megapixel is about the resolution of cheapest infrared motion capture cameras. But in the year 2000 it was the best of the best. In reality you should know the strong points of your system. Those 0.3 MP mocaps we recorded are totally usable today, especially for wide, violent motions-not so much for subtle acting when the jitter becomes quite apparent. The actors and stuntmen who can SELL the action are just as important. The suit does not magically make you into a great performer-if you make an amateurish attempt at moving in a specific way, it is often quite visible you are pretending. Stuntmen and actors experienced at staged fights will move way better than an average joe and will look much more intesting. And they can actually come up with interesting ways of beating each other up. The quality of applying the motion to the 3d character can also vary. In the previews some orcs walk in a funny way. Again, it probably does not draw too much attention in a crowd but definitely would require fixing for a character viewer will be able to see clearly.

@@Megasus-ju1mg yeah, it is easy to forget how time consuming and how heavily skill based many things in VFX are. we even forgot to mention stop motion animation.. that's wizardry.. spend so much time on frame-by-frame basis and you can't just tweak something in the end, just by reshuffling some keys and rerender the scene. and it's also interesting that stopmotion animation (e.g. something like Missing Link) can feel so grounded that you feel the weight of the characters and their movements, many times it feels more weighty than usual CG character animation. hm, the Ubisoft info is intriguing. They always had good parkour, but the connection/interpolation between animations feels somehow more janky than it used to. so they never really do any cleanup, just play with curves all the time? although thinking about it, when you mention animation layers, I guess it makes sense, because you instead just overlay your animation through that and keep some real-life jankiness, but you'll make it do what you want or for what you need. Yeah, now to think of it, this seems like a pretty standard stuff, but obviously not always. this is a bit random, but while we are at it, modern animated movies seem to me that they started to have this same-ish kind of animation. Like.. high quality and really well done animation, but it all moves in such a same manner that even characters are losing their individuality. Like, imagine if every real life actor moved in the very same manner.

Thank you so much! I'm really glad you're enjoying it 😅

That is actually true. Peter Jackson talks about it on the Appendices

Thank you!! And man, I find them in a bunnnnch of places - old articles on the internet, news reports, behind the scenes on DVDs, the DVD commentaries, interviews + my own experiences in the VFX industry to tie it all together If you're interested I've got a bunch of the online publicly available sources linked in the description - worth checking out if you wanna know more about this stuff! Thanks again for watching! I really appreciate it! 😁

I imagine they optimize it by having simpler version for sh0ts that are further away. Likevthey would do the frontlines and the point of focus in high detail and then ramp down the movement complexity. This could be done dynamically as the camera moves aswell

@@nebuli55 For movies, there are most likely no optimization as huge studios have huge render farms so they wanna render everything at highest res possible so they wanna be pixel fked by directors but they do use camera culling for shots with lots of geo.

@@xanzuls even the best computers in 1998 were an order of magnitude slower than they are now. Plenty of optimisation was needed

They probably used 3d models of different LOD (level of detail) depending on how close to the camera they are. They would also render them chunk by chunk and combine them later in compositing.

The early 2000s wasn't the Dark Ages....

@@entwistlefromthewho yeah I know, I was just wondering because even today simulations of great detail and complexity require powerful computers that were not around in the early 2000s

@@RenderBenderProductions But that obviously can't be true, because you know that late 1990s computers were performing these simulations. So either your expectation of late 90s/early 00s computers is misinformed or your understanding of how much computing power is needed to perform the task is.

render 40,000 characters at 2k resolution in less than 512Megabytes of RAM... doesn't sound too hard ;)

the only difference being video game have to render the visuals in at least 30fps in real time, but in film it has to render the image only one time to complete the scene? it's basically like making a machinima in unreal engine but for a cutscene, so you can use a lot of resources but you have to render it for a long time? right?

Mmm yea maybe, I just thought about the visual quality of blending between the different animations and captured meshes yanno? Because a large part of the volumetrically captured humans looking so good is that it's an exact 1 to 1 capture of the actor doing the movement. And once that gets tied into the crowd sim you're gonna start loosing that quality because it's gonna have to interpolate between the different captured animated states and such loosing that super "life like" appearance and start looking more CG again.

@@CGWHY I don't think the blending will be an issue, I think the end result depends much more on the quality of the performances (and their nuances) and the amount of variations.

Spoiler ⚠️ If we don’t see Paul and the fedaykin absolutely maul the emperor n his sardaukar in arrakeen in part 2 I’ll actually cry