Puppet/modular animation: taking the plunge

[Conzeit]

Turns out I'll be in non pixeled more high definition animated project so it's time I (try and) apply what we discussed at "Puppet/Modular Animation: how when and why?"

I will try to rig something that does mesh deformation in Spine or Spriter and post what I come up with. so far I've only played with the samples in each program a bit.

Has anyone given this a try?...have any tips?

[Indigo]

My only suggestion; start out as if you were doing traditional animation.  Figure out which animations you want, sketch out keyframes for them by hand.  Only then should you start dissecting which parts you'll need for the animations and how few you can get away with.  All too often I see it approached the opposite way - just making a character puppet and trying to get it to do things that it wasn't designed intentionally to do.  Just turns out looking bad that way.

[Mathias]

Also, don't forget you can combine modular with frame-by-frame drawings, selectively.
By switching out only certain parts of the animated character when extreme differences in angle occur between frames.

Here's an example spritesheet from Rayman Legends:


Notice the different options for the same body parts available to the animator.

[Conzeit]

@Indigo. I think I did exactly all the things you told me not to XD. just a little test though I'll keep that in mind when I do it for real.

Makes sense. sketch out keyframes, analyze the necesay parts for em, actually shade the parts I need.

I have a theory that's the one reason old PSX Rumblefish still looks good.

@Mathias: I've noticed that. thanks for reminding me. AND the awesome spritesheet =O

Did a little test today.
Wanted to play with mesh deformation in spine but couldnt figure out how to do that with the trial so instead I broke the model by stretching the hell out of it.



puppet is rushed, her ass breaks just by bending her legs a little :p didnt fix cause I had taken enough time anyway for a test. I probably wont use the crazy stretch technique for faking motion blur but it was good fun for a test. I also couldnt figure out how to move the whole sprite O_o. I'll try mesh deformation and on to the next software I guess.

PS: I put some more screen captures -at my blog-...if anyone's interested...? I mostly got them for my own archival reasons but who knows where you kids get your kicks now-a-days 

[Conzeit]

hum...Ive been slow on advancing on this =/

I find it weird to figure out the keyframes without uisng onionskinning and testing the timing, so I animated a rough vector stickman thing with the keyframes
https://trello-attachments.s3.amazonaws.com/53e3f6248b86f029b1b6fded/540a8716f9ac60babcbbf67e/b74ce53a660c978d5ede8aebabc98332/ArcherBasic%2BShoot.swf
the idea is to use this to identify which parts I'll need to make the actual thing. I'm still queasy, I feel like at every step I might be overcomplicating things for myself in the future

EDIT: so since this I've made some new parts for the dashing pose, I had a really hard time trying to export those vectors from flash to AI, I ended up exporting  a .png sheet from flash and drawing the vectors over it in AI. Then I had to reposition the limbs to be vertical or horizontal so as to not waste space.

anyhow I exported the parts and assembled them, then proceded to make some kinda motion for the dash


although I really liked my anticipation I was asked not to use it to make the movement as speedy as possible, so I went about animating between two keys, by using a movement graph feature in AI, it's really fun because you can make it go negative or over 100% to make anticipations and overshots and you can tweak it on each limb.

[Conzeit]

Another post on my blog, another crosspost here 

https://craftmused.wordpress.com/2014/10/20/stretch-faking-smears-to-make-cutout-animations-look-hand-drawn/

My animation workflow right now is like this
1- make keyframes with tweens disabled to make sure the posing is strong
2- tweak the curve graph between each frame to improve the timing of the tweens. Incredibly useful, you can make overshots and slow ease-in with no extra keyframes
3- On key moments individually pose each frame, even add fake smears(hand drawn motion blurs) trough stretching to give it a bit of a sakuga/anime feel.

That’s why the slash here doesnt look like cutout animation. You could just draw motion blurs and import them as parts but I’m trying not to clutter the ram of the iOS this will eventually run on.

anyone actually reading this OR the blog? I was surprised to see someone comment on it today :p

[rikfuzz]

All the poses pop in and out like a sort of poorly articulated, clockwork robot.  I'm not sure whatever you're doing with the curve graph is doing you any favours. Imagine it probably looks a lot better with plain boring flat linear tweens, moving the keys to adjust timing. 

[Conzeit]

in my tunnel vision I keyed A LOT of pop in moments in there, trying to get that anime (Yoshinori Kanada/Hiroyuki Imaishi) quality of movement that builds slowly at the beggining and skips the middle point to snaps to the end pose and the overshot or the slow ease in. So it might not even be the graph thing, just me actually keying popping in poses ( I think you might see the same thing in the SWF I linked to)

I know when I look in retrospective at something where I was trying a specific style, I see the bad parts overwhelm the good parts, so I would like to know in more detail what you think is not working, or in what moment it's more noticeable so I can see it from your perspective and have kind of a future hindsight moment =O

BTW I liked that one polygonal woman you made in flash a while ago, specially what you did with her hair in the run animation..I tried to do that but was not very succesfull

Also, I'm probably not going to do the redraw parts thing because of limitations with the atlas =/

[rikfuzz]

Flash has totally binned it's armature animation stuff, or at least changed it enough that I can't open those files again without it all being turned into keyframes, but the hair was actually done with a different kind of 'bone', that let you do arbitrary transformations on the whole shape rather than splitting into parts. (Also minor touch up in photoshop).     I had a lot of trouble trying to do it split into separate bones, too.  It looked okay, but a lot of gaps appearing at certain angles no matter how careful I was with the shapes at the pivot points. 

The popping in / out looks weird because it's stopping everything where the motion should be fastest.  It's hard to explain without a graph or examples, but instead of slow-fast-slow (spread out however you like, style-wise) for the whole movement, you have it for every pose, even though the extreme pose should be right in the middle (fast part) of the arc.  (Maybe someone else can explain this better) 

Keep it up though! 

[Conzeit]

I *think* I can see in principle what you mean. atleast partly.

I'm doing the movment by keying the timing I want, then enabling smoothing and tweaking graph. I think what you're saying is I'm applying a curve with exponential increase on every single keyfame I made and that gives the whole movement a start-stop feeling instead of an exponential increase for the whole thing.

for the most part I actually have different graphs for each frame to frame, and at the beggining I did notice the start stop thing you're talking about and try to fix it, but apparently my eye hasnt been sharp enough, so I'll keep an extra eye out for that :p

I hope that's what you meant, and what's going on. I hope it's NOT that I'm keying the momentum all wrong and it's just my sense of timing 

 thanks for the crit and the clarification

EDIT: I think I got to the crux of it. I have a sort of anticipation where he first lowers the axe, then he raises it. Then the actual attack starts and he bends his arm over his shoulder as he lurches forwards...this is the slow down you mean (?) and then increase speed again to deliver the blow,  axe comes down pretty fast. is that what you mean?

I think that might come from the fact that when I was making the key where he swings over his shoulder is when I found the method of stretching stuff so I added a few keyframes there. it does kinda ruin the momentum :p.


I actually went and made a quick tweak. I basically moved what was happening when the axe was over his shoulder to when he had his arm outstretched, moving it from the middle of the action to the beggining if I got what you meant :p. The tweak removed some anticipation frames so the jump from low axe to axe way high is insane, but is this basically what you meant?

[astraldata]

I just want to say -- I just now realized you were Camus from the old forums. xP I'm Ace from back then. I think I remember you having an avatar of a swat-looking guy stand-shooting and then kneeling to shoot. Was that you? (I know you've had your current avatar for a long time, even back then, so I may be misremembering.)

---

That said, I think what rikfuzz meant was that at all the keyframe points in your previous version seemed to rigidly "pause" for a moment around the start of each one. The "pauses" are somewhat hard to see if you stare at it too long, but it's clear that something's off due to the length of the animation and the number of points of specific articulation. In your new version, you seem to mitigate the "pauses" by removing that key where the axe is just coming up to his shoulders just before the anticipation for the strike itself. The arc of the swing is also better represented in your new animation, especially where the stretch frame is involved.

I think your biggest problem in your new one is the spacing is a bit fast from lifting the axe to over his head since you removed that keyframe. The keyframe should have been placed somewhere between the shoulders and closer to the start of the swing (rather than removed completely). Also, in your previous version, the time it took to lift the axe to the start of the anticipation for the swing greatly contributed to those perceived "pauses" to the start of the other keys, giving the appearance that it was 'stopping' at the exact moments where it should've been speeding up (of which those were mainly off due to the arc of the swing being off too much, causing the somewhat 'jerky' appearance of the attack -- though, like I said, it appears to be fixed in the new version).


---


Anyhow, I just wanted to say that I'd like to get any information whatsoever on how to go about this style of animation. I know that mesh deformation has a lot to do with how they made odin sphere, muramasa, and rayman graphics move as if they're hand-drawn, but I would mainly like to know the professional process in which they know which parts (and how to draw them) to begin with.

I've been fiddling with this for a while now, and, so far, all I've gathered (mostly from Dan Fessler) is that they do some sort of hand-drawn animation to help them decide what parts to draw for a given character. That's not enough though.

The problem is, I don't understand any decent workflow for going about that. If I've already hand-drawn the character animations (from what I understand is to help me figure out which parts to draw and how to draw them), why would I bother making them modular? We'll assume we're not doing an RPG with equippable items, etc. -- Sure, it might save *some* time, but I don't see how it would tell you what parts you need if your hand-drawn animations were as simple as possible (i.e. skeletons). To draw them as anything else, you might as well just paint in your lines at that point.

If you're going modular, with size/shape/orientation of the body parts somehow defined by the animation in a magically practical way, the time it takes to tweak the curves, precisely place each arm/foot/etc. along an arc of movement, swapping images out to resemble hand-drawn animation on certain frames, specifying the mesh for each image part, manipulating that mesh for each frame, tweaking the whole animation, adding squash & stretch, and everything else involved doesn't seem too much faster than simply drawing over your animation skeletons with nice lines and painting them in with the proper tools to achieve a superior look.

Even if a monkey could reposition the limbs, you still have to have the animation skill to know how to keep it from looking like a paper-cutout doll, so wouldn't you still need as many skilled animators on your team?

That being said, I'm simply playing Devil's Advocate here -- anyone with the knowledge to answer any of my questions here with some concrete ideas on process/workflows would be greatly appreciated by myself and others. I'd love to take the plunge into modular animation myself, but I would prefer to know how to 'dive' before I actually dive in, and the information on this subject is surprisingly sparse.

[Indigo]

I think I remember you having an avatar of a swat-looking guy stand-shooting and then kneeling to shoot. Was that you? (I know you've had your current avatar for a long time, even back then, so I may be misremembering.)

that was Pawige

[Conzeit]

nope, the guy with the avatar that kneeled to shoot was Pawige....I think he might even still have it  ...I've certainly seen it in this White pixelation. I'm sorry to say I dont really remember you.... it's amazing how awful I am at remembering people, I only remember people I talked to A LOT....  (maybe if you post some work from that time?) You can see what kind of animation I did in that "sprite" section of  my blog :p pretty much just have old stuff

I know about the rushed anticipation in the new one...it's a tentative tweak to see if I got what he meant, not an "improved" version....I'm liking that about this kind of animation, you can have something to show more easily, it's more iterative. I'm glad we got both the same thing out of Rikfuzz'es post, I'm keeping an eye out for that kind of mistake from now on.

About the vanillaware thing, people are already doing it in spine, check this out:
http://esotericsoftware.com/forum/viewtopic.php?p=16333#p16333
I dont have Spine version that can do mesh deformation so I dont know exactly how this works (does he have to manually deform everything every keyframe so it looks ok or does it naturally bend correctly?). I get by with just rotating, moving  and scaling parts, but there it is.

About how to imagine what parts you need for what poses....I'm at a loss about that myself :p it's what I least like about this whole process.
You can go and download top production quality dolls and look at them if you want. For flash there is the "My Little Pony: friendship is magic" dolls (hey, it's well animated regardless of how you like the show :p) and also the main character from Guacamelee is available to download if you buy their game. Tell me if you find more dolls...specially from Rumblefish...I still think it's one of the best examples.

I tried to do what they seem to do, which is redraw parts for every new pose, I even made that SWF where all the poses are hand drawn so I had a very spontaneous guide to start from, and I was sure I had the perfect method but I had some problems

-drawing every pose beforehand and trying to imitate it it is not very smart with spine, I do a lot of stretching in my drawing, it's so prevalent I often do it with no real purpose, so trying to recreate it in spine all the time became tedious and somewhat pointless. it did help me find the kind of timing I wanted though.
-drawing the pose with new parts, then making every part posed perpenticular so that I could export it from AI, then import it in spine and get it back in the same pose I had in AI was very time consuming, not to mention the antithesis of the spontaneity of just drawing a new frame in normal handdrawn animation 
(Indigo...how do you do it?  )
-this polygonal style requires a LOT of tweaking when you're placing parts, moving them in weird ways so the part where the model breaks doesnt show, this made for a very start and stop process every time I animated something.
-turns out I was already consuming more atlas than I should with just two poses where I used new parts for everything.

MY temporal solution was this dummy with round joints, which doesn't need any new parts whatsoever. it's not compatible with the polygon aesthetic of the game and the antithesis of what I wanted to do, but I needed to just animate, and this is the best way to do it in this technique.
When I actually add back the skin this character I will try to remove the round joints for style, I will see the need for new parts, and I may or may not draw them depending on available time and available space in the atlas.

I wanted to do EVERY cool thing ever for this technique but I may end up trying to get the most out of the most basic form of puppet animation..oh well, live and learn :p

Industry wise, the average of seconds produced with this kind of animation IS greater than fully hand drawn. Even John kricfalusy which is completely puritanic about not using tweens ended up inmersed in this. Obviously you still need professionals with a sense of timing to do it, but I would wager that the automatic tweening does help a lot even if there is still a lot of redrawing (I would imagine specially when you can tweak the movement graph...it does make for ease ins and overshots with no extra keys)

[astraldata]

Pawige! That's right -- thanks Dan! Sorry about that Conceit -- Also, thanks. It's been some years now. A lot has happened in that time for me. Its just Pawige kept me interested in pixel-based animation back in the day when I asked him how he did his avatar -- I remember getting frustrated with it to the point of almost giving up pixel art since I didn't know any of the tricks for animation in Graphics Gale at that time. Now that I know he's over on PJ, I'll just have to stop by over there and say thanks to him too!


-- Regarding Modular animation: --

As for deciding on the parts to draw, I'm glad I'm not the only one dreading that part of the process. I think if we knew how people approached this issue for film/tv, we might have a better idea of how to go about it. Those guys generally have a *lot* more stuff to draw than those of us using the process for game development. That being said, if the method Dan suggested comes from that field, I'm thinkin' maybe it's best to come up with a method more suited for games ourselves.

Tbh, I wouldn't think that it would be useful in film/tv to pre-draw your 'animations' anywhere but in the storyboarding process. I can see that being useful to determine what parts you'll need. I can see some equivalent being useful for games when thinking about it this way.

With that said however, what I see in practice with games like Muramasa and Odin Sphere (judging by the few parts sheets that are available), they appear to be animating many individual parts as if they're flipbooked, with only the upper-torso rotations, hands, feet, and stuff with follow-through (hair/drapery/etc.) containing the most frames:






Without reference, this particular character sheet is very difficult to follow due to all the drapery and overlapping parts all over the character...

BUT -- with that being said -- being 'hard to follow' visually might very well be the point for the two main character designs in Muramasa.


-- MY theory on the "hand-drawn" modular animation style is this: --

The drapery and other details are intended to hide exact angles of the bones in the shoulders, elbows, knees, and hips (in particular), which will allow for a lot greater mesh deformation options than a redraw of a forearm and shoulders, for example, might allow. By hiding the areas where the most complicated joint connections appear, you avoid having to redraw an entire section of a character's body for the sake of an obscure joint angle, allowing mesh deformations to take their place.

As you might notice, the other characters in the game tend to have very few unique animation frames -- especially NPC characters, who mostly rely on these mesh deformations.

The characters who *do* have more animation frames don't generally have very many when they've got their entire torso + arms and hips + thighs showing (i.e. boss-type characters):




A good example of a clever way of hiding the joints in an exposed set of hips is the above red demon (tengu) character -- his stomach + grass skirt hides the joints where his legs would connect with his hip-bone and his chest clearly hides one shoulder while the other foreground shoulder + arm overlaps the chest, reducing the number of images required to articulate him by doing so.

In addition to obscuring the character's joints by overlapping other parts, another clever trick was used in the demon character to smooth the skin transition from around the shoulders/arms to their attachment to the torso -- the very heavy shadows. The heavy shadowing allows anything to merge to the torso as long as it connects with the shadowed areas. That's why heavily-shadowed characters work more easily with modular animation. Concentrating heavy shadows around joint areas really helps pull transitions to other less-obscured parts off -- like the red demon's arms, for example.

That said, as far as I can tell, most cutout game characters have 2 views -- either mostly straight-on (usually with knees bent in a specific direction, like the red guy, who can use his sideways knees for both torso angles if necessary) or mostly a 2D profile view (like the guy in the below image). I've rarely seen a straight-standing character from the front that hasn't been mostly redrawn for each frame. Correct me if I'm wrong here anyone, because the girl in the scene below is one prominent example I am basing my assumption off of:



Anyway, that's where I stand right now in my current understanding of me reverse-engineering this process. Anyone with any ideas on anything I've said, especially regarding a proper workflow to determine which parts are needed for a given character design, feel free to chime in!

[Conzeit]

sigh, I still dont have the eye to see trough a sheet like that. that is very cool thank you 


I agree with you about Muramasa/Odin sphere, it's very very smooth but it's very obviuously moving only sideways, and characters look a bit robotic emotionally, so I had never looked into it.

Your points are very true tho and that means that style relies heavily on mesh deformation....if you took it away it'd be almost impossible to make it work, unlike with Dragon's Crown. (sigh...I have a lot of robed characters I'll have to find a solution for that myself :p)

I wonder if they do what I've done so far which is just to animate a regular doll, and afterwards they make a new part where it's obvious the model broke.

The reason I keep coming back to Rumblefish is they have moves where a character lunges forward rotates his torso and delivers a strong blow and trough it all the parts change accordingly and it looks very good, I dont think I really see that with anything else, even Dragon's Crown....but I wonder if it's just the low resolution that sells it to me?

[astraldata]

I added in a few additional observations in my previous post I thought were interesting points I failed to mention earlier.

--

As far as Rumblefish is concerned, much of the execution was very well done. I imagine they used the process Dan mentioned on his blog (i.e. starting out with traditional animation to attempt to see what parts they needed), which ultimately required them to pixel over quite a lot of drawings, but saving them mostly in-betweens (which likely still saved them some money since they were using pure pixel art anyway). They used a lot of redraws, but they also use some mesh deformation in there as well (you can see this more prominently in some of the breathing animations of the more rounded character forms).

That said, I keep coming back to moves like that guy's "Modular Pixel Pile Driver" (watch out -- it's PPD's arch nemesis! )

It may have been budget constraints, but it was pretty gross to look at, and a fairly solid reminder of where pixels and modular animation just don't work so well together. Though, when mixed with mesh deformation and ample redraws (like in RumbleFish), these issues can be mitigated somewhat.

There are also games (like Magicite) that treat pixel-based FFD like an art-style, and that particular instance works well due to the characters essentially being treated like single-frame "Paper Mario" styled cardboard cutouts with almost no frame-by-frame animations. In fact, the less frame-by-frame animation in these FFD style pixel animations, the better they tend to look.

This is the reason why RumbleFish works so well -- mesh deformation aside, when they *do* use modular animation techniques, they only use those for in-betweens or other subtle movements, essentially retaining that "cardboard cutout" styled FFD on the parts they use the modular techniques on. Essentially mimicking Magicite -- just on a slightly larger scale (i.e. with entire sections of a character's body on an otherwise *static* frame of animation). As long as the part itself doesn't change *while* it's being distorted, your eye won't catch it and it won't look 'off' to you. When the next frame is required, new FFD is applied to it if necessary, but in a different way, and the rest of the body tends to change too as well, leading your eye to perceive a larger change, thus leading to more fluidity in that change than could be provided by a simple redraw.

Despite its look, I'm positive RumbleFish's character sheets are much larger than Muramasa's in terms of unique parts requirements, especially with mesh deformations and FFD frames subtracted, simply because it's a fighting game. That said -- the number of unique sprites in the game are likely still less than games like the Street Fighter Alpha/Zero series.

But yeah, I do agree with you on the charm of the low-resolution hand-animated appearance of the RumbleFish games. I really do think the uniqueness of the low-res combined with the obviously modular (and modern) style stands out a bit. However, the animation technique in it is almost exactly the same as that in Muramasa -- just with the mesh deformation executed to a somewhat lesser extent.

Yet there are still many moments in RumbleFish2 that you can see that mesh deformation really well -- especially in the more 3D-looking idle frames you know are just pixel art, but your eye still tries to tell you they're 3D models somehow. With the mesh deformation overhead, it's probably still almost as much work as hand-drawing all the frames, but it does achieve a look many games don't attempt, likely for that very reason (and possibly another reason is because pixel art is also a very specialized skill).

[Conzeit]

are you sure there's FFD in Rumblefish 1 or 2? I've looked at a lot of it and havent really noticed anything that REQUIRES it, I think it's just the characters are cleverly designed to not show their joints and very subtly rotated....there might be a little bit of it in the joints...but I havent seen anything that makes me convinced.

BTW found a topic at Mugenguild where there's a character available for download.
http://mugenguild.com/forum/PHPSESSID.ojinh12aful6mjd9g9m82lmhr6/topics/request-wuwos-rumble-fish-sprite-rips-150340.0.html

the character parts have some pretty damn weird shapes. Looks like they're not designed to rotate much at all, supporting what we all thought of drawing by hand first, making parts later

This guy Felo claims to have more but I think he never uploaded em and I'm pretty sure he left Mugen...=/ (Saw a topic about that on Prime Central Station)

[Alex Sinigaglia]

Felo_Llop didn't leave Mugen. I'll ask him about the rips.


He re-uploaded them: http://mugenguild.com/forum/topics/rumble-fish-163247.msg2043527.html;topicseen#msg2043527

The rips were done by Wuwo who was a Mugenchina user.

[Conzeit]

Thanks Alex! and extended thanks to Felo. read all about who ripped it in mugenguild ( ;_; mugenchina ). I have my share of Wuwo rips too, I've been stalking mugen communities for ages for their rips :p Glad to hear Felo is ok =)

Oh these rumblefish dolls, they have such strange shapes, like they're not meant to be joints at all. I'll have to put one together to see how they work