Behind “HONEY, I AM HOME”: Building a Psychological Neo-Noir Through AI Filmmaking

HONEY, I AM HOME is an AI-generated short film about a woman whose suspicion slowly overtakes her perception of reality. What begins as a fear of betrayal turns into a psychological collapse, where imagined images become more convincing than what is actually in front of her.

The film follows a woman who believes the man she loves may be cheating. She watches, follows, imagines, and gradually becomes consumed by scenes she may or may not have witnessed. The story is not centered on whether the betrayal is real. It is about the moment when suspicion becomes an image — and the image becomes dangerous.

Developing the Story

I began with a simple premise: a woman suspects a man of infidelity and follows him. But very quickly, the story shifted away from a conventional thriller or relationship drama. I became more interested in the psychological state of suspicion itself.

I wanted the film to exist in the space between observation and projection. The viewer sees fragments: a car, a bar, a silhouette, a body, a kiss, a road, but the viewer is never fully sure whether these images belong to reality or to woman’s imagination.

The structure of the film became a gradual loss of control:

1. she watches
2. she imagines
3. she begins to believe the images
4. she reacts too late
5. the car leaves the road

The crash is not meant to function as an action sequence. It is the consequence of a delayed reaction — a fraction of a second where she is no longer fully present in reality.

Visual Language

The visual world of the film was built around cold cyan-blue light, rain-covered glass, neon reflections, dark cars, silhouettes, and overexposed roads. I wanted the film to feel like a psychological neo-noir from the 1990s: restrained, artificial, dreamlike, and emotionally distant.

The car became the central visual and psychological space of the film. It is both a vehicle and an enclosed mental chamber. She is technically in control, but internally she is drifting further away.

Mirrors, windshields, rain, reflections, and blurred light became key motifs. They allowed the film to show perception as something unstable. The world is never seen cleanly. It is always filtered through glass, light, water, or imagination.

The most difficult part of the film was the scene where she loses control of the car. I wanted the crash to feel like perception breaking down, not like a stunt. The sequence was built in small fragments: the road ahead becoming too bright, the rearview mirror stretching space behind her, intrusive images appearing in her mind, her eyes closing, her hands gripping the wheel, a late correction, and finally the black car leaving the bridge.

The key was timing. She does not lose control because of speed alone. She loses control because the images have taken over for one second too long.

Image, Error, and Control

The process of making the film started to mirror the film itself. I was making a story about a woman losing control of images, while I was also trying to control images that kept slipping away from me.

The AI would often misunderstand the emotional tone: making scenes too glamorous, too literal, too theatrical, or too polished. But some of those errors became useful. They forced me to clarify what the film was actually about.

In the end, HONEY, I AM HOME became less a story about betrayal and more a film about the danger of believing an image too deeply. It is about suspicion, projection, and the quiet violence of imagination when it begins to act on the world.

Technical Workflow

The film was built through a mixed AI workflow that combined image generation, video generation, editing, and manual visual direction.

I started by developing key still frames for the main visual moments in Midjourney: the woman inside the car, the neon-lit street, the bar, the imagined fragments of intimacy, the overexposed road, and the car falling from the bridge. I treated each shot as a separate scene with its own composition, lighting, emotional tone, and camera logic.

A large part of the process was prompt iteration. I often had to rewrite prompts several times to remove or change elements. For example, a scene that began as a woman covering her face and crying out became much quieter: a delayed reaction, a tense breath, a hand tightening on the steering wheel. The emotional direction had to be precise, because small wording changes could completely shift the performance.

For consistency, I used recurring visual rules across the prompts:

cold cyan-blue neon, rain-covered glass, dark minimal backgrounds, restrained movement, soft film grain, and 1990s neo-noir references.

The car interior, windshield reflections, silhouettes, and wet surfaces became repeated visual elements that helped connect shots created separately.

After generating the image material, I animated selected frames into short video clips. This stage required a different kind of control. Some shots worked better with almost no camera movement, while others needed only subtle motion: reflections shifting on glass, rain moving across the windshield, a slight turn of the head, hands tightening on the wheel. I avoided exaggerated movement because it often broke the psychological tone of the film.

The driving and crash sequence was assembled from multiple fragments rather than one continuous generated shot. I separated it into smaller beats: driver’s POV, the overexposed road, intrusive flashes of imagined images, her delayed reaction, the steering wheel correction, the exterior shot of the car leaving the bridge, and the impact with water. Building it this way gave me more control over rhythm and made the loss of control feel gradual rather than purely spectacular.

The final stage was editing. The film only started to make sense in the edit, where I could decide how long to hold a shot, where to interrupt reality with imagined images, and how much information to withhold from the viewer. The edit became the place where the story’s ambiguity was built.