New METAHUMANS are TOO REALISTIC in UNREAL ENGINE 5

So wouldn't it be great if we could have this level of defamation complexity but be able to evaluate in real-time.

Well, what we are now able to do is to use machine learning to compress that simulation data into a format that can be evaluated at run time.

So you are looking at our mal character with defamation driven by full muscle flesh and co from Houdini , it's running in real-time and on PS five is taking around 1/10 of a millisecond on CPU for network inference.

And as you see it here around one millisecond on GPU for morph target evaluation .

So this is a fully generalized model .

We have not just trained it on the cinematic animation we can see here.

So on the left is the nearest neighbor model with just the PC layer active .

And on the right , you can see how the additional nearest neighbor set improves the fold reconstruction.

So to help generate the optimal set of nearest neighbor poses, we have implemented a K means pose generator .

So given a set of target animations , so typically these will be your game or cinematic animations and A max pose value .

It will generate a set of poses that most efficiently cover that animation space .

We simulated our clothing for these additional poses in a Houdini and then we trained our additional nearest neighbors set using this data .

We would love for you to come and chat to us at a booth .

You can get hands on with this demo .

You can change cameras switch between muscular skeletal flesh and clothing layers .

You can pause the action and you can see the defamation result both with and without a moor form are running our guiding vision for meta human has been the democratization of complex character technologies allowing you to work faster and see the results immediately .

A character is only truly believable if its motion fidelity matches its visual fidelity .

But animating at this level is a hard task for even the most skilled studios , some of our best work leverage four D capture .

But this took specialized hardware in weeks or even months of processing time .

While me , a human creature gave you the ability to generate high quality characters .

Animating them still wasn't as easy .

This is why I'm very excited to announce a new capability to the Met a human product .

Me , a human animator , meta human animator contains the essence of our four D pipeline but optimized to run on a single machine .

It is able to use iphone as well as stereo professional systems .

And today we're going to demonstrate how it works for this .

We're going to need me our technician John Cook and just the phone .

Yeah , mel can you take your position ?

Please let me know when you're ready ?

Ok .

Ok .

And action I need performance capture to work like a mirror .

I need it to capture whether I'm acting scared or angry and sometimes all I need is a look cut .

Thanks Mel .

That was great .

Yeah , you're welcome .

Ok .

Our technician John is currently pulling mass performance from the phone onto his machine where everything will be processed locally .

We have updated our live link face mobile app to capture all data at the best resolution possible with the device .

Me human animator uses video and Deb data to convert this data into high fidelity performance animation and it can even use audio to produce convincing tongue animation .

John is currently struggling through the take to pick the section that he wants to process .

John .

Are we all good with the data ?

Awesome .

So from now on it's just a single button , click to kick off the processing which for a performance of this length will take less than a minute to convert into animation .

So Mel well that is processing .

Let me show you something else .

Oh Is that me ?

Yeah , this is what we refer to as your meta human DNA .

Cool and this is generated by the capture we made earlier , right ?

Yeah , that's right .

So from only three frames of video and that data , we can generate a rig that predicts all of your facial expressions in just a couple of minutes .

And do you only need to do this once for each actor ?

Yes , that's right .

It calibrates the solver to your face so that we can produce the performance in a way that faithful reproduces your original performance .

Sounds cool .

Yeah .

So let's check back on the , on the processing which today is on the latest C P and GP hardware from A MD .

Me human animator uses a custom epic facial solver and landmark detector .

We can interactively look at animation while it's being solved and compare it to your original performance .

So it looks like it just it's almost finished after this .

He is going to do one more pass to make the curves more stable , which is really quick .

And from here on , we can we just need to export the animation .

This takes only a few seconds and then John needs to drop it in the level and add the audio so that we can see the result .

So me me human should now be ready in the level .

Mel you excited to see the results .

I can't wait to see it .

I need performance capture to work like a mirror .

I need it to capture whether I'm acting scared or angry and sometimes all I need is a look .

Thank you .

So Mel , what do you think ?

I think it's incredible because it usually takes months between performance capture and getting any results back .

So this is blowing my mind and all of this is solved directly onto an meter framing controls .

In this case , we are using a Bespoke four D rig which we created together with N G theory for help by two .

But it's also ready to use on any medium or any other rig that follows our new Medicum standard .

Let's have a look at that .

I need performance capture to work like a mirror .

I need it to capture whether I'm acting scared or angry and sometimes all I need is a look .

So the same thing works even in stylized characters .

Thank you .

All these technologies are completely redefining our creative process as they will redefine yours .

One release me a human animator to everyone in just a couple of months .

We've got one more thing we'd like to show you .

We haven't forgotten about the need for full performance capture shoots .

What you're about to see is animation that has not been polished or edited in any way .

And it took me a human animator just minutes to process , start to finish .

Yeah .

So here's one of my favorite lines from Ninja Theory's upcoming Games saga .

Help Late two and I really hope you enjoy it and the rest of the show .

Thank you very much .

Thank you .

Yeah .

Sign Perth cut .

Take 13 .

Nice .

I see through your darkness .

Now I see three her eyes .

I will show them how to see us .

I do .

I will not appease your gods .

I will destroy them .

Good .

You like it .

Yeah .