Huanlian, or changing faces: Deepfakes on Chinese digital media platforms

Abstract

In China, deepfakes are commonly known as huanlian, which literally means “changing faces.” Huanlian content, including face-swapped images and video reenactments, has been circulating in China since at least 2018, at first through amateur users experimenting with machine learning models and then through the popularization of audiovisual synthesis technologies offered by digital platforms. Informed by a wealth of interdisciplinary research on media manipulation, this article aims at historicizing, contextualizing, and disaggregating huanlian in order to understand how synthetic media is domesticated in China. After briefly summarizing the global emergence of deepfakes and the local history of huanlian, I discuss three specific aspects of their development: the launch of the ZAO app in 2019 with its societal backlash and regulatory response; the commercialization of deepfakes across formal and informal markets; and the communities of practice emerging around audiovisual synthesis on platforms like Bilibili. Drawing on these three cases, the conclusion argues for the importance of situating specific applications of deep learning in their local contexts.

Engaging with deepfakes: a meta-synthesis from the perspective of social shaping of technology theory

“…In among the first few studies to focus on deepfakes designed not with malicious intent but for entertainment at best with monetization opportunities, Bode (2021) analyzes a deepfaked video of Keanu Reeves created by Corridor, a YouTube channel, and the role of the networked public in shaping the discourse surrounding the video and its popularity. In a departure from deepfake analysis predominantly centered on western nations, de Seta (2021) discusses deepfakes in China through the Zao app and communities of practice contributing to deepfake fabrication in the region. In this regard, the author draws attention to the impact of the Chinese technology industry on the development of deepfakes and related issues (de Seta, 2021).…”

“…In a departure from deepfake analysis predominantly centered on western nations, de Seta (2021) discusses deepfakes in China through the Zao app and communities of practice contributing to deepfake fabrication in the region. In this regard, the author draws attention to the impact of the Chinese technology industry on the development of deepfakes and related issues (de Seta, 2021). The study is significant in light of China's strict regulation of information (Guo, 2020), the Zao app's growing popularity and privacy concerns (Doffman, 2019), and China's efforts to combat deepfakes (Reuters, 2019).…”

“…Rarely, such motivations may also be associated with blackmailing or cyberbullying (Sabrina, 2022) or the execution of spear-phishing on high-potential targets (Ghazi-Tehrani and Pontell, 2021). Intriguingly, deepfakes may not always be constructed with malicious intent, but rather as a pastime or leisure tool (de Seta, 2021; Widder et al. , 2022).…”

“…, 2021; Bode, 2021; Eelmaa, 2022; Gamage et al. , 2022; Newton and Stanfill, 2020; de Seta, 2021; Widder et al. , 2022).…”

“…In this context, pro-deepfake arguments imply that deepfakes provide a platform for creative expression (Ali et al. , 2021; Bode, 2021; de Seta, 2021; Widder et al. , 2022), a means to engage with the celebrity being deepfaked (Bode, 2021; Popova, 2019) and an opportunity to showcase technological expertise and hone skills (Newton and Stanfill, 2020).…”