Epic SloMo and Twixtor

[Kemalettin Sert]

The one above saying that it looks like sh** really needs some kind of (visual) therapy or something, Sorry..

It looks just amazing to me.. but one thing that I noticed immediately in those shots after watch them, is that the camera is completely steady, while is the subject moving across the scene, is where Twixtor really does a pretty good job. Don't know about pan movements with the camera thou.

therapy? thanks for thinking me
that footage looks all blurry and lack of single detail to me..blame my eyes

 

[Justin Chin]

There is plenty of warping going on because of the cross motion against the mud. You generally don't notice it that much with fluids (though I saw it in this example) because you might generally forgive it. In some instances you can see whole patches of a splash moving in conjunction, and not on each droplets own expected path. You can also see warping in the feet and background. There's plenty of it, and some of it is quite odd.

As many have said here, you have to take in the fact what you're trying to accomplish, test them, and see if it will work for you. I have had some great results with a parkour shoot I had but we made sure we shot in a way that would give us the best results, i.e. not a lot of background detail. The flatter the better.

Also, I personally wouldn't use most of, "The Beauty of Mud" shots on a big screen. That warping will be really apparent.

 

[Justin Chin]

And I'll just add, that I find the examples really amazing. I just would find some of it hard to use for the big screen (and on big productions - but that's not what we're talking about).

 

[Björn Benckert]

The cinespeed / twixor / motion estimate timewarp errors when they are used as "a cool thing"....

Those where the days: http://www.youtube.com/watch?v=aRYokc3VBC4

 

[Sergio Perez]

I'm a fan of twixtor, have used it a lot, some in broadcast work, and in my experience Twixtor loves "single" no foreground and background multiple moving objects shots. Liquids is also a no no. The morphing gets terrible in these situations. But if you get a uniformed motion and separated subjects, it can look pretty stunning. But it all ends up very trial and error...

Two examples here:

Twixtor at 10% from Canon 7D footage (STart playing at about 00:30 seconds)

https://vimeo.com/32508174

Red Epic at 300 FPS (No twixtor) (This one has less super slow motion on the racing but towards 00:30 there's a sequence that looks similar to the above twixtored shots)

https://vimeo.com/54079531


You wont notice much difference here, but what you haven't seen is the amount of shots I've "failed" with twixtor to achieve the results in the 7D footage that I got " straight out of the camera" with the epic...

 

[Uli Plank]

100% coincident with my experience regarding bot Epic and Twixtor.

 

[Jean-Claude Benet]

yeah with too much going on in foreground vs background twixtor can give some gnarly artifacts, but I'll say it's an awesome tool in the right situation. it all just depends on what you need but I do think it has it's place. would like a twixtor test like that done with 5K footage.
since we're on that page, here's a cool little video I saw a while back shot on 60d and twixtored:
https://vimeo.com/27612305

 

[paulherrin]

yeah with too much going on in foreground vs background twixtor can give some gnarly artifacts, but I'll say it's an awesome tool in the right situation. it all just depends on what you need but I do think it has it's place. would like a twixtor test like that done with 5K footage.
since we're on that page, here's a cool little video I saw a while back shot on 60d and twixtored:
https://vimeo.com/27612305

about 30 seconds in or so there's some very strange rippling warp on his face while his knee crosses the frame - perfect example of the limitations. now, i'm perfectly confident you could pick a shot like that apart and do some paint/roto and make it work, even for the big screen perhaps, but by the time you've done all that on multiple shots, why didn't you just rent what you needed in the first place and be done with it?

 

[paulherrin]

just as a reference point, the epic (no pun intended) super-slomo van falling shot shown over and over in 'inception' used interpolation... but they knew what they were doing. you can hide particle motion artifacts with real/cg particles mixed in...

[searching for article...]

found it: http://if.com.au/2010/12/15/article/Inception-behind-the-Oscar-short-listed-VFX/XMBSWQQVES.html
GREAT interview/article... must read.

 

[Björn Benckert]

about 30 seconds in or so there's some very strange rippling warp on his face while his knee crosses the frame - perfect example of the limitations. now, i'm perfectly confident you could pick a shot like that apart and do some paint/roto and make it work, even for the big screen perhaps, but by the time you've done all that on multiple shots, why didn't you just rent what you needed in the first place and be done with it?

One day phantom flex rent is usually about 2.5k USD. I wonder if the guy doing the flips are willing to pay that kind of money or if he rather save those and get little ripples in his face when jumping

That film is a great example where it work kind of fine, clear blue sky behind, no patterned clothing and where the editing is done to taste and so on... The scenario differs quite a bit if you are going to shoot a little miniature that needs to explode over cranked or such... For those one shot kind of shots, I recommend not to say "we fix it with twixor..." for obvious reasons

 

[paulherrin]

One day phantom flex rent is usually about 2.5k USD. I wonder if the guy doing the flips are willing to pay that kind of money or if he rather save those and get little ripples in his face when jumping

haha, twixtor away on youtube... the overall effect is intact. i love the tool, simple and quick (and sometimes dirty). it's certainly opened up a lot of creative possibilities, albeit at the expense of purity. there are definitely shots you couldn't get otherwise (and shooters/editors that wouldn't be able to experiment). all i'm saying is there ain't much that beats the real thing. the only thing that comes close is a lot of work (with knowledge/wisdom).

 

[Dmitry Burenok]

Regardless of shooting situation is always better to use real high speed camera for over cranks like 1000FPS and up. So generally phantom does it w-a-a-a-a-a-y better than any twixtor. Twixtor cheats it, sometimes - nice and sexy, sometimes - totally useless. And twixtor only works more or less with settle camera or really "plain" backdrop. And it has to be really fast shutter 45 degrees or even faster. From other side it is much more chance to get better results when you slow down from 120FPS to 600FPS than from24FPS to 120FPS

 

[Russ Fill]

Some of the issues Ive had with Twixor makes me think I must have taken the software back in 74 a few times before music festivals and now I have flash backs when I don't get the contrast or subject just right in frame when using it now. I think I would rather use the real thing in fps and not suffer the side effects.

 

[Bob Gruen]

The key point to understand is, Twixter does not add any information, it just interpolates between the existing data.

This is exactly wrong; Twixter is interpolating existing frames to make new frames, thus adding information. Before the process the frames do not exist, after the process they do exist. This thread is really asking the question: How well does Twixter create the new frames from the existing ones, and what is the tradeoff?

The answer is that it is totally subject driven and subjective to the viewer, so your best bet is to shoot a few very accurate test shots and take them through the post process before production so you know how you want to do it, or shoot it 3 or 4 different ways during production so you have choices in post.

Looking at the Vimeo post for example: In the very first shot there is some warping in the flying mud that looks fake. We all know that things fly through the air in arcs, caused by momentum and accelerations due to gravity and influenced by air/wind. The movement of the flying mud in the first shot is too simple so the warping jumps out. Other shots where the mud is more complex probably have warping too, but the complexity of the mud hides it. It is a bit counter intuitive in that regard, which is why a highly accurate test shot (just do it) is really the only way to know for sure.

As is usually the case the best question to start with is: How will the material be exhibited? If it will be a small format on the web AND I have enough lighting, I'd just rent the high speed camera and shoot the subject at the desired frame rate. If it is meant to be a theatrical release I'd test to see how much I could get out of combining Epic footage with the Twixter process in post. I would lean toward preserving resolution over accuracy, but we all have our own personal tolerances in that regard. For example, some people think the posted Vimeo footage looks like...

 

[Pablo Moreno]

This is exactly wrong; Twixter is interpolating existing frames to make new frames, thus adding information. Before the process the frames do not exist, after the process they do exist. This thread is really asking the question: How well does Twixter create the new frames from the existing ones, and what is the tradeoff?

The answer is that it is totally subject driven and subjective to the viewer, so your best bet is to shoot a few very accurate test shots and take them through the post process before production so you know how you want to do it, or shoot it 3 or 4 different ways during production so you have choices in post.

Looking at the Vimeo post for example: In the very first shot there is some warping in the flying mud that looks fake. We all know that things fly through the air in arcs, caused by momentum and accelerations due to gravity and influenced by air/wind. The movement of the flying mud in the first shot is too simple so the warping jumps out. Other shots where the mud is more complex probably have warping too, but the complexity of the mud hides it. It is a bit counter intuitive in that regard, which is why a highly accurate test shot (just do it) is really the only way to know for sure.

As is usually the case the best question to start with is: How will the material be exhibited? If it will be a small format on the web AND I have enough lighting, I'd just rent the high speed camera and shoot the subject at the desired frame rate. If it is meant to be a theatrical release I'd test to see how much I could get out of combining Epic footage with the Twixter process in post. I would lean toward preserving resolution over accuracy, but we all have our own personal tolerances in that regard. For example, some people think the posted Vimeo footage looks like...

Thanks for the physics class man but I think you are wrong about something, Twixtor is not creating something that is not in the image already, Twixtor analyses all the frames, the very first and the very last prior to do any interpolation, that's why is so slow, because it has to put all that information on Ram first, and no, not all objects flying in the air make arcs immediately, because it depends on the mass of the object and its acceleration and many other factors. That's why Twixtor is not very good when the image is too blurry, because it cannot "see" what's in there..

 

[Nick Morrison]

Thanks for the physics class man but I think you are wrong about something, Twixtor is not creating something that is not in the image, Twixtor analyses all the frames, the very first and the very last prior to do any interpolation, and no, not all objects flying in the air make arcs immediately, because it depends on the mass of the object and its acceleration and many other factors. That's why Twixtor is not very good when the image is too blurry, because it cannot "see" what's in there..

I think you are both right, and looking at the same thing from different directions. On the one hand, yes Twixtor is only working with what's there. It's not inventing new materials. But as it is ADDING new information, filling in the blanks as it spaces things out, so it is in effect also ADDING data. It's not actually capturing fluids suspended in air at 1000fps. It's guessing at what it looks like. And interpolating frames that try and emulate the real thing (ie perfect motion, arc, texture). Capturing these qualities in camera will always be more succesful, because there is no guess work.

However if the "math" is simple, it stands to reason Twixtor can do a pretty bang up good job.

It should be seen as a budget solution, one without the reliability of a professional, in-camera approach.

 

[Pablo Moreno]

I think you are both right, and looking at the same thing from different directions. On the one hand, yes Twixtor is only working with what's there. It's not inventing new materials. But as it is ADDING new information, filling in the blanks as it spaces things out, so it is in effect also ADDING data. It's not actually capturing fluids suspended in air at 1000fps. It's guessing at what it looks like. And interpolating frames that try and emulate the real thing (ie perfect motion, arc, texture). Capturing these qualities in camera will always be more succesful, because there is no guess work.

However if the "math" is simple, it stands to reason Twixtor can do a pretty bang up good job.

It should be seen as a budget solution, one without the reliability of a professional, in-camera approach.

Thanks Nick, but is also true what somebody posted before; you get better results going from 120fps to 2000fps than 24fps to 120fps.. So an Epic camera and Twixtor will be a very good combo, even if you have the budget for a Phamtom one..

 

[paul schefz]

is there a resolution limit with twixtor? i can't find any info on their site...

 

[Nick Morrison]

Thanks Nick, but is also true what somebody posted before; you get better results going from 120fps to 2000fps than 24fps to 120fps.. So an Epic camera and Twixtor will be a very good combo, even if you have the budget for a Phamtom one..

Makes sense. I just wouldn't promise a client 1000fps and rely exclusively on Twixtor. Personally I'd rather promise 300fps at 2K. And if Twixtor can deliver some stunning 1000fps as a bonus, then so be it.

Question: what's better. 120fps 5K twixtored to 1000fps. Or 300fps 2K twixtored to 1000? Presumably the 300fps is already so slow-mo, that it's easier "math" for Twixtor, right?

Love your guys thoughts on this.

 

[Jeff Kilgroe]

therapy? thanks for thinking me
that footage looks all blurry and lack of single detail to me..blame my eyes

I've got your back, dude. I also think it looks like shit.

It's usable in a pinch for what they did there, but I would never pitch that as the first option for such shots on any paid professional job. Ironically, Twixtor actually does quite well with fluids and things that tend to contort and distort as they move since it does a better job of fooling our brains into seeing more than what is there. Actually the last commercial I did with slow-mo water used mostly CG. We shot on EPIC and while the FX budget could have probably been reallocated to get us a Phantom, we tried setting up the shots anyway. We wanted it for reference when doing the FX and could never get the water to fly and break just the way the director wanted anyway.

As many others here have pointed out, a lot depends on the type of motion and detail in the shot. Fluid splashes, particulates in the air, etc.. are some of the most forgiving subjects because they naturally create lots of points to track and are amorphous. When you get multiple planes of motion overlapping one another, it falls apart. You can see that in the very first retimed shot in the example video. Look at all the muddy water splash to the left side of the person. IMO, I think that video could have been made better if they had shot it with a faster shutter -- so much of it is a blurry mess.

Anyway, Twixtor and other similar tools can be powerful if used right. And they are often used in conjunction with high-speed image capture to extend the effect. A 2X increase in frame rate, or creating a single intermittent frame following each real frame, usually works great. It's enough to get creamy, chocolatey looking liquids for commercial work and the EPIC. If you need something running at 1200fps, that's a tall order if you need it to look right. A lot depends on the subject matter and there are always options, especially for film production where it doesn't have to be "real" where it can be faked. 120fps to 1200fps means you have to magic up 9 frames out of thin air in between every real frame you have. That almost never looks right, no matter what you do. However, for the needs of certain jobs, it may be good enough.