Cutting Wings: Human and AI Collaboration in Short Film Production and the Representation of Truth

Cutting Wings combines science fiction, crime, and suspense to ask how an image comes to be accepted as truth. In its fictional world, ECHO reconstructs past events from memories retained by physical objects. Its institutional reputation makes these reconstructions appear objective. The engineer's discovery of conflicting fragments interrupts that certainty and places him between obedience to an official account and the pursuit of an uncomfortable alternative. The dramatic conflict therefore concerns the human interests surrounding the system as much as the apparent reliability of its images. The project connects this narrative question with the conditions of its own production. Research on latent diffusion describes how conditioned image synthesis can produce detailed images, while video diffusion extends generation into moving sequences [1,2]. These developments provide a technical context for AI filmmaking, but a convincing appearance does not establish that an event occurred. Chesney and Citron [3] discuss how synthetic audiovisual media can undermine trust in recorded evidence. Cutting Wings approaches this concern through fiction, using generated images to question the authority granted to images. Our creative objective was to give this problem an ordinary, recognizable setting. Modern workplaces, restrained lighting, and a cold, low-saturation palette anchor the speculative device in familiar institutions. Figure 1 presents an AI-generated production image of the ECHO product demonstration. Its orderly display, formal presenter, and seated press audience visualize the system's public authority. This framing helps explain why the engineer's doubt becomes a conflict with the surrounding institution rather than a simple technical troubleshooting exercise.

We used a practice-based approach centered on developing and assembling the film. The production materials consisted of the script, shot plans, character and environment references, generated video clips, synthesized voices, and edited sound. ChatGPT assisted script and storyboard development, and AI image tools supported concept design. Image-to-video generation, including Kling AI in the production records, animated selected still images. AI voice synthesis and AI-assisted sound and music work supplied audio material. Editing and color grading brought the outputs together. The workflow in Figure 2 summarizes these stages and the points at which human review informed revision [4]. Preproduction began with the dramatic function of each scene. We translated narrative information into shot descriptions specifying the subject, action, framing, and environment. Character references established recognizable faces and clothing, while environment references defined the workplace and the ECHO device. Reusing these references gave later generations a stable starting point. Prompt revisions focused on observable requirements, such as preserving the protagonist's dark work uniform, the device's proportions, and the relative positions of desks and screens. This preparation linked visual continuity to the story's needs. For moving images, we divided scenes into shorter shots instead of asking a generator to complete an extended sequence of actions. A shot with a clear primary action was easier to evaluate and revise. Restrained gestures, fixed compositions, and limited camera movement also suited the protagonist's hesitant behavior. We reviewed generated material for changes in identity, clothing, hands, props, and spatial relationships. The production records treated legible interface text as a separate concern: essential information could be added in post-production rather than relying on generated lettering. Audio and picture were coordinated during editing. Synthesized dialogue required adjustments to timing and pauses so that speech corresponded with the character's performance. Sound effects and background music were refined alongside the sequence's rhythm. Color grading addressed differences between independently generated shots. Review focused on whether an action remained understandable across cuts and whether the visual treatment supported the intended emotional state. This was a qualitative creative process, with no controlled comparison or statistical measurement of production efficiency.

We assembled generated images, moving shots, voices, and sound into the completed short film. Its main creative result is a connection between the production method and the narrative theme. The film uses AI to construct a believable fictional world in which an AI system's apparently authoritative reconstruction can be questioned. The ECHO demonstration in Figure 1 provides a visual account of institutional confidence; the engineer's discovery introduces uncertainty into that confidence. Their relationship makes the system's credibility an element of the dramatic conflict. We used birds, feathers, and scissors as recurring symbols. The bird represents innocence and freedom, the feather suggests memory and truth, and the scissors express the forces that restrict freedom and justice. These associations connect the title to the engineer's limited ability to act or speak. The workflow in Figure 2 shows how those artistic decisions precede and guide generation, then remain relevant during editing. The resulting visual associations depend on the filmmakers' selection and arrangement of material, rather than an assumption that AI independently understands the symbolism.

Maintaining character consistency was one of our clearest production difficulties. A convincing isolated image could still fail as part of a sequence if the face or clothing changed. Fixed character references and repeated prompt refinement addressed this problem by making continuity an explicit production requirement. Preserving the protagonist's dark work uniform illustrates why this matters: a costume change can distract attention from the character's decision and make adjacent shots seem to belong to different scenes. Reference use supported continuity, but did not remove the need for human inspection. Natural movement required a related change in shot design. Generating long sequences placed several demands on the same clip, including performance, object handling, and camera movement. Dividing scenes into smaller units allowed the filmmakers to review each action and build the sequence's rhythm through editing. This approach suited the film's restrained thriller style, in which a pause or a small gesture can carry dramatic weight. It also created an editorial responsibility: individually usable shots still had to preserve screen direction, object placement, and the causal order of events. Dialogue synchronization demonstrates how work continued beyond generation. We spent substantial time coordinating AI voices with character performances. Speech timing influences how viewers read hesitation, doubt, and pressure. A voice that moves too quickly through a pause can weaken the emotion suggested by the image. Adjusting dialogue and shot duration therefore served both technical alignment and characterization. Similar care was necessary for interface lettering, because distorted text could obscure information that the audience needed to understand the evidence. The film's use of synthetic images also creates a productive tension. Its story challenges the equation of visual coherence with truth, while its production seeks coherence so that viewers can follow the story. In this context, cinematic continuity is an artistic construction; it does not certify a factual reconstruction. ECHO's ability to read object memories belongs to the fictional premise. The film offers a way to discuss trust and institutional pressure, and should not be interpreted as a demonstration of real forensic technology or its accuracy. The project supports a practical conclusion about human-AI collaboration. AI supplied material for visualization and performance, while we determined which images belonged together, what information remained visible, and how the final sequence expressed doubt. Our experience suggests opportunities for quicker experimentation and lower production demands, although we did not collect timing, cost, or audience-study data to measure those benefits. The findings are therefore limited to this creative case. Future work could document rejected outputs and revision time, and examine whether viewers understand the conflicting evidence and symbolic imagery as intended. Cutting Wings integrates generative tools into a short-film workflow while maintaining a clear question about the authority of reconstructed images. Fixed references, short shot generation, and manual audio and editorial adjustments formed the main responses to production difficulties. The experience places human judgment at the center of both filmmaking and the story's ethical concern: a persuasive image still requires someone to question how it was produced and what it leaves out.