Critical Daily    Critical practice in graphic design

The development in Generative Adversarial Networks (GANs) brought many new tools for creating synthetic visuals, sounds, or texts. The most famous neural network to generate photorealistic visuals is StyleGAN [1] from NVIDIA. When it is trained on a particular category dataset, for example, photos of human faces (Flickr-Faces-HQ dataset [2]), it learns to generate new, unseen human faces of people that never existed. What new possibilities does this tool bring, and are there any concerns about its wide use?

I focus on applying synthetic media in audiovisual production and critical experimentation with ML/AI models that generate synthetic media content. From the position of a visual creator, I delve into the new ways of creating visuals in the dialogue with AI. From the designer's role, I question AI tools' future applications and try to imagine plausible scenarios that synthetic media might bring. Combining these two roles together, I aim to establish a conscious practice and spread awareness on conflicting moments that might occur.

StyleGAN Anonymizer is an anonymizing tool that shows a practical application of generated synthetic media. Using AR face filters, as we know from Instagram or Snapchat, a random face generated by the StyleGAN neural network is placed on the location of a person's face on whom the phone's camera is pointing. In this way, it is possible to make photo documentation of events with increased protection of the privacy of the persons involved or to anonymize photographs additionally. The filter uses face recognition to adapt the synthetic face to the facial expression of the person being photographed and may, in the future, act as a substitute for dispersing or blackening the face.

At the same time, this project reveals the unpreparedness of social platforms for synthetic content, as Facebook did not approve the upload of this filter to Instagram due to a violation of policy no. 2.11. This policy talks about banning the use of photos of people's faces in filters. The synthetic faces that the filter contains never belonged to any existing human being and are, therefore, in my view, as fictional as drawings or illustrations. But their photorealism creates the illusion of representing reality, and the algorithm (or administrator) can't distinguish them from real photographs of human faces.

Therefore, the filter is available only at Snapchat, which does not have such a strict policy. Snapchat turns StyleGAN Anonymizer to another level using the virtual camera app (Snap Camera [3]) during video calls. From now on, you can enter video calls incognito but still wear a legit face. That brings us closer to imagining the synthetic reality we are slowly entering and reevaluating the value of our (physical or synthetic) identities.

[1] https://github.com/NVlabs/stylegan
[2] https://github.com/NVlabs/ffhq-dataset
[3] https://snapcamera.snapchat.com