AUTOMATIC PERSONALIZED AVATAR GENERATION FROM 2D IMAGES

Smart overview of the Invention

The patent application describes a system for generating personalized avatars from 2D images. This is achieved by a generative component on a virtual experience server, which creates a 3D mesh of the user's face from these images. The 3D mesh is then converted into a polygonal mesh, which is fitted and rigged onto a head portion of an avatar model. This automated process simplifies the creation of avatars, eliminating the need for manual design efforts.

Technical Context

The invention addresses the complexities involved in creating avatars for online platforms, such as gaming environments. Traditionally, users needed to manually design 3D meshes using specialized software, a process requiring significant expertise. The described system automates this process, allowing users to generate avatars without needing advanced skills. This approach leverages machine learning to handle tasks like mesh shaping, texturing, and rigging, which are traditionally labor-intensive.

Functionality

The system utilizes machine-learning models to generate feature vectors from input images and user-provided prompts. These vectors help create a 3D mesh that is rigged onto an avatar model for display in virtual environments. Users can input images of their face, and the system provides guidance on image storage and use, ensuring compliance with local regulations. In jurisdictions where facial data storage is restricted, alternative methods like stock images or text prompts are used.

User Interaction

Users can customize their avatars with various styles and features by responding to prompts. These prompts might include descriptors like "caricature," "cartoon," or references to popular culture and personal traits. The system supports diverse customization options, allowing users to emphasize or alter specific facial features, or to mimic styles from movies or TV series. This flexibility enables the creation of unique and personalized avatars.

Technical Implementation

The generative component is trained to produce accurate feature vectors and 3D meshes using models such as Generative Adversarial Networks (GANs). Training involves creating a computer-generated face representation and converting it into a 3D mesh. The system may use a combination of conditional and unconditional models, with options for user selection of starting faces. This trained model facilitates the automatic generation of avatars, streamlining the process for users.