This paper proposes a novel and generic attacking strategy called Deformation-Constrained Warping Attack (DeCoWA) to enhance adversarial transferability across different model genera. DeCoWA uses elastic deformation to augment input examples, preserving global semantics while enriching local details. It introduces an adaptive control strategy to constrain the strength and direction of the warping transformation, ensuring the consistency of global semantics in augmented samples. The method is applied to various modalities, including images, videos, and audio, and demonstrates superior transferability compared to existing attack methods. The approach is evaluated on multiple tasks, including image classification, video action recognition, and audio recognition, showing significant improvements in attacking systems with different model genera. The results indicate that DeCoWA can effectively hinder the performance of models such as Transformers on tasks like image classification, video action recognition, and audio recognition. The method is also visualized using Grad-CAM to show how it affects the attention mechanisms of different models. The paper highlights the importance of addressing the challenges of cross-model genus attacks and presents DeCoWA as a promising solution for improving adversarial transferability across diverse model architectures.This paper proposes a novel and generic attacking strategy called Deformation-Constrained Warping Attack (DeCoWA) to enhance adversarial transferability across different model genera. DeCoWA uses elastic deformation to augment input examples, preserving global semantics while enriching local details. It introduces an adaptive control strategy to constrain the strength and direction of the warping transformation, ensuring the consistency of global semantics in augmented samples. The method is applied to various modalities, including images, videos, and audio, and demonstrates superior transferability compared to existing attack methods. The approach is evaluated on multiple tasks, including image classification, video action recognition, and audio recognition, showing significant improvements in attacking systems with different model genera. The results indicate that DeCoWA can effectively hinder the performance of models such as Transformers on tasks like image classification, video action recognition, and audio recognition. The method is also visualized using Grad-CAM to show how it affects the attention mechanisms of different models. The paper highlights the importance of addressing the challenges of cross-model genus attacks and presents DeCoWA as a promising solution for improving adversarial transferability across diverse model architectures.