AI RESEARCH

MOLM: Mixture of LoRA Markers

arXiv CS.LG

ArXi:2510.00293v2 Announce Type: replace-cross Generative models can generate photorealistic images at scale. This raises urgent concerns about the ability to detect synthetically generated images and attribute these images to specific sources. While watermarking has emerged as a possible solution, existing methods remain fragile to realistic distortions, susceptible to adaptive removal, and expensive to update when the underlying watermarking key changes.