Text this: Toward Adaptive Unsupervised and Blind Image Forgery Localization with ViT-VAE and a Gaussian Mixture Model