Last updated: Jun 5, 2023

Unreal Engine 5.2 uses machine learning to compress simulation data into a format that can be evaluated in real-time, allowing for greater complexity in animation.

It also introduces a new feature called MetaHuman Animator, which can capture high-fidelity performance animations using video and Dev data from a phone or other devices.

The creator challenges traditional beliefs by exploring new ways of designing game characters in the engine and receives positive feedback for their work.

  • Unreal Engine 5.2 compresses simulation data using machine learning to allow for greater animation complexity.
  • MetaHuman Animator captures realistic performance animations using phone data and can even use audio.
  • A nearest neighbor model and pose generator are implemented to optimize animation space.
  • Simulated clothing and additional nearest neighbor sets are trained to improve animation quality.
  • The engine can be used for full performance capture shoots, producing high-quality animations in minutes.
  • The creator challenges traditional beliefs and receives positive feedback for their work.

Machine learning to compress simulation data 00:08

  • Unreal Engine 5.2 is able to use machine learning to compress simulation data into a format that can be evaluated in real-time.
  • This allows for a level of deformation complexity to be evaluated in real-time, which was not possible before.
  • The simulation data is compressed into a format that can be evaluated at runtime.

ML character with decimations 00:15

  • The ML character shown on the video has decimations driven by full muscle flash and class emotion from Houdini.
  • It is running in real-time on PS5 and is taking around one-tenth of a millisecond on CPU for network inference.
  • As shown in the video, it takes around one millisecond on GPU for morph targets evaluation.

Nearest neighbor model 00:36

  • The fully generalized model shown in the video is not just trained on cinematic animation.
  • On the left is the nearest neighbor model with just the PCA layer active, and on the right, an additional nearest neighbor set improves the fold reconstruction.
  • To help generate the optimal set of nearest neighbor poses, a k-means pose generator is implemented.
  • Given a set of target animations, it generates a set of poses that most efficiently cover animation space.

Simulating clothing and training additional nearest neighbor set 01:11

  • Clothing is simulated for additional poses in Houdini.
  • The additional nearest neighbor set is trained using the simulated clothing data.

New capability - MetaHuman Animator 02:16

  • MetaHuman Animator is a new capability to the MetaHuman product.
  • It contains the essence of the 4D pipeline but optimized to run on a single machine.
  • It is able to use iPhone as well as stereo professional systems.

Performance capture using phone 02:45

  • MetaHuman Animator uses video and Dev data to convert data from a phone into high fidelity performance animation.
  • It can even use audio to produce convincing tongue animation.

MetaHuman DNA 03:59

  • MetaHuman DNA is generated by the capture of only three frames of video and data.
  • It can generate a rig that predicts all of the facial expressions in just a couple of minutes.
  • It calibrates the solver to the face, producing a performance that faithfully reproduces the original performance.

Full performance capture shoots 06:46

  • MetaHuman Animator is not just for stylized characters, it also works on full performance capture shoots.
  • The animation shown in the video has not been polished or edited in any way and took MetaHuman Animator just minutes to process from start to finish.

Creating realistic spider 07:47

  • The video shows the process of creating a realistic spider using the Unreal Engine 5.2.
  • The creator demonstrates how to see and appreciate the details of the spider, which is a common fear for many people.
  • By creating a realistic spider in the game engine, it is possible to understand and appreciate the natural beauty of the creature.

Challenging traditional beliefs (473 - 476)

  • The creator challenges traditional beliefs by saying "I will not appease your gods, I will destroy them".
  • This statement reflects the idea of moving beyond traditional beliefs and exploring new ways of thinking.
  • In the context of the video, it may be referring to the traditional way of creating game characters and exploring new ways of designing them in Unreal Engine 5.2.

Positive feedback 08:11

  • The creator receives positive feedback, saying "good you like it".
  • This feedback indicates that the work being done is appreciated and well-received.

The importance of diversity 08:30

  • The creator emphasizes the importance of diversity by using the word "foreign".
  • This may be a reference to the idea of including diverse characters in the game engine and exploring new ways of thinking about representation.
  • The use of the word "foreign" suggests that there is a need to go beyond traditional ways of thinking and explore new perspectives.