AI Image Editing with Iterative Text-Guided Consistency Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial intelligence-based image editing technologies struggle to maintain image consistency before and after action editing, undermining the editing effect.
Innovation Solution
Perform noise addition on an object image to obtain a noisy encoding vector, encode an action instruction text to get a first action text encoding vector, iteratively denoise the noisy vector using the text encoding vector, update the encoding vector based on the difference between the denoised image and the original, perform fusion processing on the encoding vectors, and finally denoise the noisy vector to apply the action to the object image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If action editing technology is used to edit images, then the image editing effect is improved, but the image consistency before and after editing deteriorates
Solution Approach 1:
The patent applies preliminary action by performing text encoding on the action instruction before the denoising process, creating a text encoding vector that guides the subsequent image generation. This ensures the editing action is properly formulated before being applied to the image, maintaining consistency while achieving the desired editing effect.
Solution Approach 2:
The patent implements feedback through iterative denoising processes where the denoising result of one round serves as input for the next round. This feedback mechanism allows the system to progressively refine the image while maintaining consistency with the original image content and the action instruction, resolving the contradiction between editing effect and image consistency.
Data Source
AI summary
An artificial intelligence-based image processing method includes: performing noise addition on an object image to obtain a noisy image encoding vector; performing text encoding on an action instruction text to obtain a first action text encoding vector; denoising the noisy image encoding vector based on the first action text encoding vector to obtain a first action image; updating the first action text encoding vector based on a difference between the first action image and the object image to obtain a second action text encoding vector; performing fusion processing on the first action text encoding vector and the second action text encoding vector to obtain a fused action text encoding vector; and denoising the noisy image encoding vector based on the fused action text encoding vector to obtain a second action image.