High-resolution image editing is increasingly vital in professional workflows, yet existing diffusion-based models are limited to resolutions below 1K due to quadratic attention complexity and high memory demands. A common workaround uses a two-stage pipeline: editing at low resolution followed by independent super-resolution. However, this approach suffers from information divergence, where hallucinated details contradict the original high-resolution (HR) source, and texture degradation, manifesting as over-smoothed or over-sharpened artifacts.
To address these issues, researchers from Hugging Face introduce EditBridge, a diffusion bridge framework for efficient ultra-high-resolution editing. Unlike conventional diffusion that regenerates from noise, EditBridge formulates refinement as structured data-to-data translation from the low-resolution (LR) edited result to its HR counterpart, explicitly conditioned on the original HR source to preserve authentic details.
To efficiently incorporate HR source guidance, the framework introduces a prior-guided block-wise sparse attention mechanism. This exploits semantic correspondence from the first-stage editing to constrain cross-image interactions to spatially aligned regions, significantly reducing computational overhead.
Extensive experiments demonstrate that EditBridge achieves high-fidelity editing with superior perceptual quality at resolutions up to 4K. It delivers a 3.6–8.4× speedup at 2K and enables practical 4K editing in just 61 seconds, marking a significant step toward real-world ultra-high-resolution image editing.