US20260045012
2026-02-12
Physics
G06T11/60
The patent application introduces a method for generating images using generative artificial intelligence (AI) based on user inputs. The process involves receiving a request and a descriptive prompt from the user, then selecting an appropriate machine-learning model from a predefined set to generate the desired image. This approach aims to enhance the user experience by providing a more tailored and efficient image creation process.
Generative AI has the capability to create images from text prompts, but existing solutions often produce unrealistic results with limited applications. Users typically need to switch between different tools to achieve their desired edits. The current systems lack an integrated approach, leading to fragmented user experiences and inconsistent outcomes due to the specialization of machine-learning models.
The described method involves selecting the most suitable machine-learning model based on the type of image requested and the user's prompt. It can also involve generating a rewritten prompt to refine the selection process. This methodology allows for the generation of various types of images like stickers or avatars, and can include additional functionalities such as animations and clothing modifications for avatars.
The system comprises processors and computer-readable media that execute instructions to carry out the image generation process. It includes receiving user requests, selecting appropriate models, and generating images that meet user specifications. The method also supports iterative refinement of prompts, allowing users to modify their requests for improved output images.
This innovation addresses the limitations of current generative AI systems by offering a more cohesive and user-friendly solution. By intelligently selecting machine-learning models and integrating various functionalities, it provides a seamless workflow for creating and editing digital content, enhancing the creative potential and user satisfaction.