Conducting social research on images uploaded by people on the Internet can be very complicated, as it is necessary to ask their permission in order to respect their privacy. This slows down or even makes many studies impossible. With the Camouflage project, however, the Polytechnic University of Turin has attempted to make use of Gan and AI to produce anonymous images that, starting from the original ones found online, result as modified versions of the latter, different just enough to make them unrecognizable, while maintaining comparable meaning.To test the goodness, but also the limitations of this system, the computer engineers who created it contacted a group of semiologists who are experts in the study of images, and in particular those of faces, given their affiliation with Massimo Leone's Erc Facets (Face Aesthetics in Contemporay E-Societies) project. They asked them to compare the meaning of the original ones downloaded from the Internet, those produced by the Camouflage algorithms, and those made by some recommendation systems comparable to them. By means of the Delphi method (Linstone, Turoff, 1975), setting the reflections from the theoretical foundations of sociosemiotics (Marrone, 2001; Ferraro, 2012; Santangelo, 2013) and identifying a number of problems to which it is imperative to provide an answer (they will be faced by Barbotto, Gramigna, Marino and Voto), this research has produced some reflections that demonstrate the interest of collaboration between computer engineers and visual semiologists, within the nascent field of the so-called "synthetic data" (Jordon et al, 2024).