US20260148513
2026-05-28
Physics
G06T19/20
The patent application discusses a system for generating personalized avatars from 2D facial images. Utilizing a generative component on a virtual experience server, the system converts these images into a 3D mesh data structure. This structure is then transformed into a polygonal mesh, which is automatically fitted and rigged onto an avatar model, creating a personalized virtual representation of the user.
This technology pertains to virtual experiences, specifically focusing on the automatic creation of avatars from 2D images. It addresses the complexities of manually designing avatars, which traditionally require extensive effort and specialized software skills. The invention simplifies this process by automating the creation of 3D meshes and integrating them into avatar models.
The system employs machine learning models to generate feature vectors from user-provided images and text prompts. These vectors are used to create a 3D mesh, which is then rigged onto an avatar for use in virtual environments. Users can customize their avatars by responding to prompts about desired features, styles, and attributes, allowing for a wide range of personalized virtual representations.
Users can upload images depicting their faces, which are temporarily stored for mesh generation purposes. The system provides options to proceed with or without images, ensuring compliance with local regulations on data usage. Images are deleted after use, and face-based generation is restricted to jurisdictions where it is legally permissible. Users can also delete generated meshes and associated data upon request.
A generative component, trained to produce accurate feature vectors and 3D meshes, underpins the system. Training involves a combination of 2D generative components and 2D to 3D transformation models, potentially using Generative-Adversarial Networks (GANs). This training allows the creation of avatars from both user images and predefined styles, enhancing the flexibility and personalization of the avatar generation process.