Visual Media Generation With Prompt-Based Image Quality Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing visual medium generation technologies struggle to accurately produce visual media with consistent and desired image quality levels, as existing models lack effective methods for training and fine-tuning based on image quality evaluation.
Innovation Solution
A method and apparatus that utilize a loss function to train visual medium generation models by determining the difference between evaluated and indicated image quality, fine-tuning generative models, and generating training data to improve image quality through prompts, including image quality evaluation and comparative evaluation data generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If existing visual medium generation models are used, then visual media can be generated, but the image quality consistency and accuracy cannot be ensured
Solution Approach 1:
The patent implements a feedback mechanism by introducing an image quality evaluation model that assesses generated visual media and compares it against prompt-indicated quality levels. The loss function uses this evaluation feedback to guide the optimization of the visual medium generation model, continuously improving image quality consistency and accuracy through iterative refinement based on evaluation results
Solution Approach 2:
The patent changes the training parameters by incorporating image quality evaluation scores as additional loss function components. Instead of only training on basic generation accuracy, the model now optimizes for specific image quality parameters (such as clarity, detail, and aesthetic quality) that are evaluated by the evaluation model, thereby improving both consistency and accuracy of image quality
2Ease of manufacture
If visual medium generation models are trained without image quality evaluation, then training process is simpler, but image quality control is insufficient
Solution Approach 1:
The patent introduces an image quality evaluation model as an intermediary component between the visual medium generation model and the training process. This intermediary evaluates the quality of generated images and provides feedback through the loss function, enabling precise image quality control without requiring complete redesign of the training process
Solution Approach 2:
The patent segments the training process into distinct components: the basic generation task and the image quality evaluation task. By separating these functions and combining their loss functions, the system maintains the simplicity of the original generation model training while adding quality control capabilities through a modular evaluation component
Data Source
AI summary
A processor-implemented method includes obtaining a plurality of prompts indicating image quality with different levels, and generating a plurality of visual media of the same content corresponding to the respective prompts by applying the obtained prompts to a visual medium generation model, wherein the visual medium generation model is trained based on a loss function related to a level of image quality evaluated for an output visual medium and a level of image quality indicated by an input prompt.