CAMOUFLAGE is an innovative image anonymization method developed by a computer science team led by Lia Morra on behalf of the ERC FACETS group headed by Massimo Leone at the University of Turin, Italy, using Latent Diffusion Models (LDMs). Unlike previous Generative Adversarial Network (GAN)-based techniques, it anonymizes entire images — including faces, bodies and backgrounds — while retaining key analytical attributes such as facial features and spatial scene composition. This approach generates synthetic data that is valuable for social research as it provides anonymized yet insightful datasets. The paper presents semiotic comments on the chimerical nature of these facial data, which are necessary to preserve the epistemological level without discarding the feasibility of the research, leading to a re-evaluation of the established dichotomies at stake (e.g. image/face, figurative/plastic, permanent/ephemeral, physiognomy/pathognomy, text/enunciation, type/token).CAMOUFLAGE aims to anonymize images to prevent identification while maintaining a level of analyzability that retains meaningful content (i.e. the derived image should retain a meaning “comparable to the original”). This can only be achieved by focusing on supra-individual, typological features (including facial expressions and gestures) to the detriment of individual, idiosyncratic features. This ensures that the anonymized images maintain a meaningful level of information without revealing specific identities. If we need to preserve meaning when anonymizing, it cannot refer to the face depicted in the original image, because conversely, we simply should not anonymize it. On the contrary, we can try to preserve as much meaning as possible from the image that shows this face.