T-Edit: Triple-Branch Diffusion Anchoring for Consistent Editing

While Multimodal Diffusion Transformers (MMDiTs) have achieved remarkable success in high-fidelity generation, maintaining semantic faithfulness and structural consistency during the image editing process remains a fundamental challenge. DiT-based editing is primarily limited by cumulative drift and semantic leakage induced by new textual conditions. To address these challenges, we propose T-Edit, a training-free framework that formalizes consistent editing as a trajectory anchoring process. T-Edit explicitly decouples the inversion, reconstruction, and editing trajectories, leveraging the reconstruction branch as a structural reference to achieve real-time compensation for deviations in the latent manifold. To further reveal the internal regulation mechanism of DiTs, we analyze the spatio-temporal heterogeneity of their layer-wise structural sensitivity and accordingly propose a Dynamic Vital Layer (DVL) localization mechanism based on information energy. Furthermore, addressing the asymmetry of textual perturbations in the frequency domain distribution, we introduce a frequency-aware strategy based on tensor Singular Value Decomposition (t-SVD) to anchor (TA) high-rank structural components. Experiments show that T-Edit achieves state-of-the-art performance in both semantic alignment and structural fidelity, and can be seamlessly extended to multi-step editing and video scenarios, providing a new perspective for understanding and controlling the internal stability of DiTs.

Text-guided image editing allows users to modify an image using natural language, such as adding an object, changing an action, or altering the style. However, modern diffusion models often change parts of the image that should remain unchanged, especially after multiple editing steps. This paper introduces T-Edit, a training-free method that keeps the editing process anchored to the original image while still following the new text instruction. The method identifies which internal layers of the model are most important for preserving structure and uses frequency-aware guidance to reduce unwanted changes. Experiments show that T-Edit produces edits that better match the target prompt while preserving the original layout, identity, and temporal consistency in both images and videos.