We propose CHIMERA, a zero-shot image morphing framework that improves structural stability and semantic coherence through adaptive feature reuse and shared-anchor text conditioning. We also introduce GLCS to evaluate global harmonization and local transition smoothness. Experiments show improved morphing quality across diverse diffusion backbones without retraining.

Jeahun Sung is a PhD candidate in the CMLab at Chung-Ang University, advised by Prof. Jihyong Oh. Jeahun received a B.S. in Mathematics, with a double major in Computer Engineering, from Dong-A University and an M.S. in Imaging Science from Chung-Ang University. Jeahun’s research focuses on low-level vision, spanning image refinement, restoration, super-resolution, morphing, and video demoiréing. A particular research interest is leveraging generative priors to address real-world degradations and improve the quality of visual content.

Dahyeon Kye is a PhD student in the CMLab, advised by Prof. Jihyong Oh. Dahyeon Kye’s research focuses on generative models for visual content, spanning image and video generation, editing, and restoration. Additional research interests include 3D/4D generation, world simulation, and interactive generative models that allow users to interact with and shape dynamic visual worlds.