DeHate: A Stable Diffusion-based Multimodal Approach to Mitigate Hate Speech in Images (arXiv)

In addition to distorting public discourse, the proliferation of harmful online information presents serious obstacles to preserving a safe digital environment. To address this, the researchers provide a multimodal dataset that has been specially designed to detect hate in digital content. The creative use of watermarked, stability-enhanced, stable diffusion techniques in conjunction with the Digital Attention Analysis Module (DAAM) is at the heart of the suggested methodology. It plays a crucial role in identifying the hateful parts of pictures and producing thorough hate attention maps, which are then used to blur the offensive areas out of the picture. The dehate shared effort includes the release of this data set. Additionally, a vision-language model called DeHater is introduced, which is intended for multimodal dehatification tasks.

https://www.arxiv.org/abs/2509.21787