Reflection piece by NYU PhD student and ExpressiveAI reporter Clone Wen
Two years ago, in November 2023, I wrote an ExpressiveAI post on “The Epistemic Anxiety of AI Visualizations”. Two years later, the epistemic anxiety is now unfolding across social media: major platforms require AI-generated videos to be labeled, yet this has not stopped the collective impulse to scrutinize every clip and debate on whether it is AI-generated or a deepfake. AI has, inevitably, entered the workflows of film and media production.
The fastest changing domain is AI animation. Last month, Professor Sun Lijun (Vice President of the Beijing Film Academy and Director of its Animation School) came to NYFA to speak about the pedagogical shift in Chinese film schools. As part of a major state-funded research initiative, AI-generative animation has been introduced as a foundational course at both the undergraduate and graduate levels. A showreel of Professor Sun’s Asian ink-wash animated film and his students’ work demonstrated the effective use of AI in producing 5-minute shorts. Another example of artist using AI in animation is LuYang’s DOKU the Creator (2025, 61 min), which I discussed in last month’s post, AI Reincarnation: Lu Yang’s DOKU Series. LuYang completed this feature-length independent animated film in one year—something that would normally require longer work from an entire Hollywood studio crew. AI has brought an unprecedented liberation of labor in animation production, and may well lead to a new wave of independent animated filmmaking.
The second domain is AI in non-fiction films. The epistemic anxiety of AI might invite documentary and other documentary-related genres to return to the foreground. This fall, I heard two NYU professors mention in public talks how AI can be beneficial for non-fiction films: “AI can be a good thing.” In visual anthropology-related practices such as oral history, essay film, animated documentary, and ethnofiction, AI can be a good thing—it replaces traditional high-cost shooting-production with simple AI generations at home. AI has the potential to attract top writers into the field of film—anthropologists, journalists, novelists—who can now author their own films. Apart from AI video tools, they may use AI music software to compose OST, and finally use AI to design posters, realizing their long-held dream of making films themselves.
Last but not least, AI has made a huge contribution to film subtitling. AI can recognize, generate, and translate subtitles. In the United States, international films frequently suffer from unprofessional and undervalued subtitle translation, making it difficult for audiences to understand. Vice versa, a film or TV series that is distributed internationally generally needs Spanish subtitles for Latin America; French, German, and Italian subtitles for Europe; and, in Asia, Japanese and Chinese subtitles, etc. AI speech-to-text and AI translation have facilitated regional film festivals that suffered from a lack of human translators and subtitle glitches. Several directors, artists and producers I met this fall are also using AI subtitling tools to help their films’ international screenings and distribution.
There are still lingering concerns of AI filmmaking—Invisibility of Latent Space in The Epistemic Anxiety of AI Visualizations: “The repeated iterations replace a poisoning dataset are witnessing the formation of giant feedback loops of AI-generating AI, which will further solidify the poisoned areas of invisibility.” This loop of AI-generating AI is happening right now: GPT is in its fifth generation, Sora in its second, Gemini in its third, Midjourney in its sixth, and so on. Small glitches are accumulating in poisoned areas of invisibility, solidified through each iteration, inside data center giants: in the liminal space between prompt and generation, in the brief moment of AI “thinking,” what exactly does AI see—and what does it choose, selectively, or deliberately, to hide?
Expressive AI 