EDITBRIDGE: Towards Faithful and Efficient Ultra-High-Resolution Image Editing
EditBridge targets the failure point in 4K edits: keeping source details intact after a low-res edit.
The paper frames common two-stage editing as lossy: low-resolution diffusion edits followed by separate super-resolution can hallucinate details or damage texture. EditBridge instead treats the second stage as a guided translation from the edited low-res image to a high-res result, conditioned on the original high-res source. Its sparse attention limits cross-image matching to aligned regions to reduce compute. The authors report editing up to 4K, with 3.6-8.4x speedups at 2K and 4K editing in 61 seconds. HF Daily Papers' note
The paper frames common two-stage editing as lossy: low-resolution diffusion edits followed by separate super-resolution can hallucinate details or damage texture. EditBridge instead treats the second stage as a guided translation from the edited low-res image to a high-res result, conditioned on the original high-res source. Its sparse attention limits cross-image matching to aligned regions to reduce compute. The authors report editing up to 4K, with 3.6-8.4x speedups at 2K and 4K editing in 61 seconds. HF Daily Papers' note
score 4