Training-Free Reward-Guided Image Editing via Trajectory Optimal Control

Researchers have developed a new training-free framework for reward-guided image editing using diffusion models. The approach treats image editing as a trajectory optimal control problem, allowing for better preservation of source image content while enhancing target rewards compared to existing methods.

• →New framework enables training-free, reward-guided image editing using diffusion and flow-matching models.
• →The method formulates editing as a trajectory optimal control problem with iteratively updated adjoint states.
• →Approach significantly outperforms existing inversion-based training-free guidance baselines.
• →Framework achieves superior balance between reward maximization and source image fidelity.
• →Method avoids reward hacking while maintaining semantic content preservation.