DirectEdit: Step-Level Accurate Inversion for Flow-Based Image Editing

With recent advancements in large-scale pre-trained text-to-image (T2I) models, training-free image editing methods have demonstrated remarkable success. Typically, these methods involve adding noise to a clean image via an inversion process, followed by separate denoising steps for the reconstruction and editing paths during the forward process. However, since the reconstruction path is approximated using noisy latents from mismatched timesteps, existing methods inevitably suffer from accumulated drift, which fundamentally limits reconstruction fidelity. To address this challenge, we systematically analyze the inversion process within the flow transformer and propose DirectEdit, a simple yet effective editing method that eliminates the inherent reconstruction error without introducing additional neural function evaluations (NFEs). Unlike most prior works that attempt to rectify the inversion path, DirectEdit focuses on directly aligning the forward paths, enabling precise reconstruction and reliable feature sharing. Furthermore, we introduce a preservation mechanism based on attention feature injection and multi-branch mask-guided noise blending, which effectively balances fidelity and editability. Extensive experiments across diverse scenarios demonstrate that DirectEdit achieves efficient and accurate image editing, delivering superior performance that outperforms state-of-the-art methods. Code and examples are available at https://desongyang.github.io/Directedit.

Modern AI image editing tools can change an image according to a text instruction, such as replacing an object, changing the weather, or applying a new artistic style. However, a common problem is that these tools often change parts of the image that should remain untouched. For example, when asked to change only one object, they may also alter the background, distort the original structure, or lose important details. This paper introduces DirectEdit, a new method for making text-guided image editing more accurate and reliable. The key idea is to help the AI model better preserve the original image while performing the requested edit. Instead of allowing small errors to build up during the editing process, DirectEdit carefully keeps the editing path aligned with the original image at each step. This makes it easier to keep unchanged regions stable and to reuse accurate details from the source image. DirectEdit also identifies which parts of the image should be edited and which parts should be preserved, so that local edits, background edits, object additions, and style changes can be handled more precisely. This makes it useful for creative design, digital media, and other applications where users want controlled and faithful image edits without needing to train a new model.