Deepfaking Keanu: YouTube deepfakes, platform visual effects, and the complexity of reception

Abstract

On July 14, 2019, a 3-minute 36-second video titled “Keanu Reeves Stops A ROBBERY!” was released on YouTube visual effects (VFX) channel, Corridor. The video’s click-bait title ensured it was quickly shared by users across platforms such as Facebook, Twitter, and Reddit. Comments on the video suggest that the vast majority of viewers categorised it as fiction. What seemed less universally recognised, though, was that the performer in the clip was not Keanu Reeves himself. It was voice actor and stuntman Reuben Langdon, and his face was digitally replaced with that of Reeves, through the use of an AI generated deepfake, an open access application, Faceswap, and compositing in Adobe After Effects. This article uses Corridor’s deepfake Keanu video (hereafter shorted to CDFK) as a case study which allows the fleshing out of an, as yet, under-researched area of deepfakes: the role of framing contexts in shaping how viewers evaluate, categorise, make sense of and discuss these images. This research draws on visual effects scholarship, celebrity studies, cognitive film studies, social media theory, digital rhetoric, and discourse analysis. It is intended to serve as a starting point of a larger study that will eventually map types of online manipulated media creation on a continuum from the professional to the vernacular, across different platforms, and attending to their aesthetic, ethical, cultural and reception dimensions. The focus on context (platform, creator channel, and comments) also reveals the emergence of an industrial and aesthetic category of visual effects, which I call here “platform VFX,” a key term that provides us with more nuanced frames for illuminating and analysing a range of manipulated media practices as VFX software becomes ever more accessible and lends itself to more vernacular uses, such as we see with various face swap apps

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

“…In this regard, additional research on the intricacies of media literacy and its impact on deepfake detection may be worthwhile to take into consideration. 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.…”

“…, 2022). We also observe instances of online channels creating deepfakes to generate revenue while providing entertainment value (Bode, 2021; Gamage et al. , 2022).…”

“…, 2021; Ghazi-Tehrani and Pontell, 2021) and erosion of societal values, ethics and morals in the long run (Ali et al. , 2021; Bode, 2021; Eelmaa, 2022; Gamage et al. , 2022; Newton and Stanfill, 2020; de Seta, 2021; Widder et al.…”

“…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).…”

Aberrant AI creations: co-creating surrealist body horror using the DALL-E Mini text-to-image generator

“…First identified by roboticist Masahiro Mori (1970), in recent years the concept of the uncanny valley has been applied to computer-generated effects in films or deepfake videos circulating on social media that are considered unsettling. Deepfake videos use AI to graft the hyper-realistic physical (and sometimes vocal) likeness – typically of celebrities or politicians – onto a surrogate body and without the figure’s consent (Bedingfield, 2022; Bode, 2021; Fletcher, 2018), with uncanny results for viewers who question their authenticity. Similarly, the hyper-realistic computer-generated human–cat characters in the musical Cats (2019) were critiqued for being too creepy (Lodge, 2019).…”