A new trend has taken social media by storm, with users creating videos that parody the shortcomings of AI-generated content. This phenomenon, which originated on Chinese platforms TikTok and Bilibili, has now spread to X (formerly Twitter) in the US, engaging users with its humorous take on AI video fails.
The parodies specifically target image-synthesis videos that attempt to morph subjects into other people or objects in physically impossible ways. Users replicate these anomalies without special effects, positioning their bodies in unusual ways as unexpected objects appear on camera.
In a real sense, what these videos represent are the current limitations of AI-generated videos, despite the considerable investments from tech giants like OpenAI and Runway. While these companies have strived to improve their models with more training data and computational time, the results have fallen short of convincing realism, especially when depicting scenarios outside their training datasets.
OpenAI’s Sora and Runway’s Gen-3 Alpha, both unveiled in early 2024, depict the cutting-edge of AI video generation. These models can create realistic scenes when closely matching examples found in their training data. However, when tasked with generating content beyond their narrow set of constraints, the results often veer into absurdity, which provides easy content for imitators.
Some famous examples of AI video failures include a viral clip from last year attempting to show actor Will Smith eating spaghetti, which became the subject of widespread mockery. Smith himself joined the parody trend earlier this year, poking fun at the bizarre depiction. Another infamous case was an AI-generated beer commercial from May of last year, created using Runway’s earlier Gen-2 model, which showcased AI’s limitations in producing coherent and realistic content.
While we still have a long way to go before AI use in video production can become seamless, that future is still on the horizon. As machines become more adept at generating realistic content, the line between AI-generated and human-created content will likely become so blurred that such distinctions will become meaningless. Really, it’s only a matter of time.