New METAHUMANS are TOO REALISTIC in UNREAL ENGINE 5.2
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.
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.
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.