A presidential motorcade in Dallas 1963
In collaboration with Williams Cole, we tested the ethical and security risks of using generative AI tools like Runway AI and MidJourney to create synthetic archival media, assessing how malicious actors or state adversaries might exploit this technology to fabricate historical events for propaganda.
Artificial intelligence has reshaped how we document, interpret, and even fabricate history, providing tools that blur the line between restoration and distortion. My recent collaboration with scholar and filmmaker William’s Cole explores these complexities, specifically addressing the ethical and security implications of AI-driven synthetic archival media. Our project examines how these tools, while powerful for restoration, risk misrepresenting the past, creating an unsettling capacity for manipulation that could redefine our trust in archival history.
Generative AI produces strong familial resemblance for “synthetic sisters” circa 1880
A Technical Approach to Authentic Fabrication
The technical process of creating convincing synthetic media for archival purposes requires a deep understanding of film and video characteristics across multiple eras. Generative AI often seeks perfection, but the authentic aesthetic of historical media comes from imperfections—irregularities rooted in the technology of the time. To achieve this, we’ve developed a range of techniques to bring the “flaws” of analog and early digital media to life, creating a media experience that feels as close to genuine as possible.
For film stock emulation, a significant focus has been on replicating film grain profiles specific to the era and type of stock. We apply unique grain types to reflect the characteristics of various films, with attention to the nuanced, almost organic variations that were standard across stocks. Gamma curves are carefully adjusted to match the specific look of historical film stocks, while color science—including color transformations and conversions—adds another layer of accuracy. Scratches, gate weave, sprockets, and flash frames evoke the mechanical imperfections of celluloid, capturing the look of film starting to roll within the magazine, with flash frames that provide subtle cues to the film’s physical nature.
When it comes to analog video, the challenge lies in reproducing artifacts intrinsic to early electronic media. We incorporate compression artifacts, high-contrast gamma curves, and simulated blanking intervals with time codes to evoke the era. These blanking areas, usually unseen but present along the edge of analog video frames, contain those characteristic blips and beeps, which further add to the realism. Lens and in-camera effects, such as star filters, specific zooms, and era-typical lensing styles, add the stylistic touches common to specific periods, as well as imperfections like whip pans and image skewing from sensor readout issues.
For digital video, we layer in compression artifacts and explore low dynamic range qualities inherent to digital media from the 80s, 90s, and early 2000s. Digital systems from these eras captured images differently, creating a unique “video-esque” look often defined by limited contrast ratios and imperfect color reproduction. Each period of digital video has its own specific compression and dynamic range characteristics, which we emulate to ensure a seamless historical feel.
Bystanders running away from shooter on the grassy knoll circa 1963
Our project serves as both a technical and ethical investigation into AI’s role in historical media. As these tools become more advanced, they present new opportunities to preserve and, paradoxically, distort history. It’s our hope that this work will raise awareness about the authenticity of archival materials in the digital age, underscoring the importance of preserving truth amidst the possibilities of artificial recreation.
A street news interview in Manhattan circa 1993
Tintype portrait complete with “scanned” frame circa 1880s
The Surreal Encounter with Synthetic History
Working with generative AI to create synthetic archival media feels almost like stepping into a surreal, alternate past. The experience goes beyond simple fabrication; it’s as if I’m “synthesizing” a reality that never was, peering into moments that could have existed but didn’t. There’s an uncanny depth to the faces that emerge on the screen—faces of people who seem so real, like characters I could have known. Occasionally, the AI generates a resemblance between these synthetic figures, like siblings or family members who share distinct traits, giving them an eerie familial bond. It’s an emergent property of the technology, yet it brings a strange, haunting familiarity that I never expected. Seeing these almost-family members, created by code yet evocative of lived experiences, stirs something deep within—a mixture of fascination, disquiet, and even a quiet melancholy, as if I’m encountering fragments of a forgotten world that I’ve somehow helped bring into being. It’s both sadly nostalgic and unsettling, a reminder of how powerfully AI can blur the line between reality and imagination.