SYSTEM AND METHOD FOR LEARNING AND COMMUNICATING IMPLICIT STYLISTIC PREFERENCES FROM HISTORICAL USER INTERACTION DATA IN TEXT-TO-IMAGE PROMPT ENGINEERING

Smart overview of the Invention

The patent application outlines a system and method designed to incorporate stylistic preferences into an image generation model through user interaction data. The system generates multiple images from user prompts, allowing a second user to select preferred images. These selections are displayed in a node tree diagram, visually representing the relationships between different sets of images and highlighting preferred choices.

Technical Field

This invention operates within the realm of text-to-image generation, employing machine learning to create related images for design projects. It emphasizes capturing client stylistic preferences, which are crucial for designers and creative professionals using generative AI tools to align their work with client expectations effectively.

Methodology

The method involves generating image sets for each user prompt, establishing relationships between these sets, and allowing a second user to choose preferred images. These selections are weighted positively or negatively based on user feedback, influencing the node tree diagram's structure by adding or removing images accordingly. This process aids in refining design directions with fewer iterations by learning implicit stylistic preferences.

User Interface

The user interface plays a pivotal role in displaying the node tree diagram and facilitating interaction between users. It allows users to visualize the relationship between visual concepts through semantic distance calculations and chronological ordering. This interface is versatile, supporting collaborations between various user pairs such as designers and clients or employees and managers, enhancing project development efficiency.

System Components

The system comprises a user interface and a style and prompt learner, receiving inputs like text prompts and feedback to generate a searchable history of design inspiration. A text-to-image model processes these inputs to produce corresponding images, which are then displayed for user interaction. This setup enables a dynamic adaptation to client preferences, streamlining the design process by learning from historical interactions.