In diffusion models, inversion methods learn structured noise that allows for the reconstruction of the original image or for faithful editing with minimal alterations. While most methods try to improve the inversion trajectory to find better structured noise, they often hinder editing flexibility. The recent RF-Inversion method proposes a controlled ODE to guide the reverse process, but it still faces instability and requires extensive tuning. To address this issue, we propose a Dynamic Look-ahead Controller (DLC), a training-free, plug-and-play module with surrogate targets that are applied on the existing rectified velocity field. We derive DLC from a dynamic optimal control formulation and prove that it reduces to a linear–quadratic regulator with a time-varying target, recovering fixed-target RF-Inversion as a special case. We further introduce a control strength scheduler that largely eliminates the necessity of manual parameter tuning. Through extensive experiments and ablation studies on the LSUN and SFHQ datasets on stroke-to-image inpainting and identity-preserving face editing tasks, we demonstrate that our method achieves state-of-the-art performance.