Title: High-Resolution Editing Can Be Achieved Through Low-Resolution Operations

Abstract: The ability to generate high-resolution content has become a pivotal concern within the fields of computer vision and graphics. As images remain the primary medium for visual communication, creating content that accurately reflects user intent demands robust and controllable mechanisms for manipulating high-resolution visuals. Yet, current methodologies are largely constrained to low-resolution environments, usually capping out at 1K resolution. This paper addresses this gap by defining the task of high-resolution image editing and presenting a test-time optimization framework designed to tackle it. Our approach applies patch-wise optimization to high-resolution source images, complemented by a fine-grained detail transfer module and an innovative synchronization strategy to ensure coherence across different patches. Comprehensive experiments demonstrate that our technique yields high-quality edits, thereby advancing the capabilities of high-resolution content creation.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC