At the intersection of AI and film, generative tools are fundamentally changing our relationship with the moving image. This excerpt from my book, AI Cinematic Realism, dives into those shifts to help define the future of cinematic truth.
From its very beginning, cinema has carried the burden and the promise of realism. When the Lumière brothers screened their first short films in 1895, audiences famously ducked as a train seemed to barrel toward them from the screen. The astonishment was not simply about movement or novelty—it was about a new kind of presence, an image that appeared to capture reality itself.
This foundational moment crystallized a tension that has haunted film theory ever since: is cinema defined by its ability to replicate the real, or by its power to construct illusions?
Across the 20th century, the most influential film theorists framed realism as cinema’s core identity. Siegfried Kracauer, writing in the aftermath of World War II, described film as the “redemption of physical reality,” a medium whose greatness lay in its capacity to record the world in its contingency and detail. André Bazin, in his famous essay “The Ontology of the Photographic Image,” argued that the photographic basis of film anchored it in a unique relationship to reality, granting cinema a privileged link to truth.
For both Kracauer and Bazin, the photographic image was more than a representation: it was an indexical trace of the world, a record of “what has been.” This indexicality grounded cinema’s claim to realism. The moving image was, in their view, a window onto reality, even when stylized or fictionalized.
Of course, realism has never been a simple matter. Theories of montage, mise-en-scène, and genre demonstrate that realism is constructed through choices of framing, editing, and style. Yet the photographic base—the fact that light from the world left its imprint on celluloid—provided the foundation on which debates about realism could unfold. Even when cinema embraced fantasy or spectacle, it did so in dialogue with its anchoring in the real.
The Rupture
Digital technologies complicated this foundation, but they did not erase it. The shift from celluloid to digital cameras in the late 20th century sparked concern that cinema’s indexical bond was weakening. Still, digital images were usually tethered to captured reality: a digital camera, like a film camera, recorded light from the world, even if pixels replaced grain. Computer-generated imagery (CGI) expanded possibilities for spectacle, but it was largely folded into live-action footage, maintaining some continuity with indexical realism.
Artificial intelligence, however, marks a more profound rupture.
When an AI generates an image, it does not begin with light bouncing off the world. Instead, it produces new images out of patterns in data, trained on vast datasets of prior images. The AI-generated image is not a record of “what has been,” but a synthesis of what could be made to appear.
There is no lens, no scene, no performance—only prompts and probabilities. The image doesn’t refer back to something that existed; it simulates the appearance of having done so. This is a fundamental ontological break from both analog and digital filmmaking. Even CGI-heavy productions still rely on motion capture, green screens, or photoreal textures grounded in real-world inputs. With AI, the image may be entirely synthetic from the outset.
Generative models such as diffusion networks or large language–vision architectures do not need to point a lens at the world to create an image. A face, a street, a battlefield, or a dreamscape can be conjured without ever existing before the lens. The AI image is not a trace of light but a statistical synthesis—an image that owes its existence to patterns in data rather than the presence of the world.
From Indexicality to Plausibility
This rupture places AI images in a different ontological category than both celluloid and digital cinematography. For Kracauer or Bazin, this would amount to a radical dislocation. Realism, in their terms, depended on cinema’s privileged bond to reality; AI unmoors the image from that bond.
This shift reframes how realism operates. In place of photographic truth, we get plausibility: images that look like they could have been recorded, even though they weren’t. We’re no longer responding to traces of the real—we’re responding to how convincingly those traces can be mimicked. Realism becomes less about fidelity to the world, and more about stylistic coherence, narrative fluency, or affective believability.
This change doesn’t just affect how we view media—it affects what we trust. When the image no longer guarantees any contact with reality, we enter a new terrain: one in which realism is manufactured not by optics, but by algorithms.
This raises the central provocation: if cinema’s identity has always been tied to realism—grounded in indexicality, framed by philosophical debate, and contested through style—what happens when AI severs or reconfigures that bond?
It forces us to accept that AI Cinematic Realism is not a contradiction in terms, but a prompt to reexamine the foundations of what realism has always been assumed to mean. We are witnessing the emergence of cinematography that isn’t filmed at all. Instead of manipulating a physical environment, creators input prompts or assemble data to generate images from scratch.
This is the horizon of our new field: a field that acknowledges the rupture introduced by AI, while exploring the continuities, perceptions, and debates that make its images feel real.
If these ideas resonate with you, the full manifesto explores the intersection of film theory, practice, and ethics in greater detail.
Joni Gutierrez, Ph.D. 
Leave a comment