Overview

Cyborg Trailer is a film trailer project I created through an AI-assisted creative workflow. In my previous AI-related projects, I had never really explored film-oriented content—especially not a movie trailer, which demands strong character consistency, fast pacing, and a more cinematic sense of shot design. I chose this subject because I wanted to challenge myself while also testing the current creative limits of generative AI as of April 2026.

​

To begin, I used Nano Banana Pro to design the main character, Cyborg—a red robotic protagonist. Through repeated revisions and refinements, I eventually developed a complete set of three-view character references. During this process, I also used Tripo Studio (Tripo 3D) to generate a 3D model based on those character views. I then brought the model into Maya, where I set up basic lighting and rendered more accurate reference images from multiple angles. My goal was to create stronger visual references that could later be fed into AI video-generation tools such as Seedance, helping preserve character consistency as much as possible across different shots.

I also experimented with generating environmental reference images using ChatGPT-Image 2, but the results were not ideal for this particular project. In the end, I returned to Nano Banana to develop the environment references, creating a cyberpunk-inspired modern cityscape with a more suitable visual style. Through ongoing discussions with ChatGPT, I gradually refined the overall cinematic style, camera language, and shot design until I arrived at the direction I wanted for the trailer.

​

Once the visual direction was clear, I generated two to three video outputs for each shot in both Sora 2 and Seedance 2.0. From these different generations, I selected the strongest moments and edited them together in Premiere Pro to build the final trailer. I then created the logo animation manually in After Effects, and finally used Topaz Video to upscale the finished piece.

​

This project was both a creative experiment and a technical test. It showed me that, with enough preparation and strong reference images, AI tools can already support a surprisingly cinematic workflow. At the same time, it also made the current limitations of generative AI very clear—especially in projects like trailers, where pacing, continuity, and character consistency are far more demanding than in shorter or more impressionistic pieces. For me, Cyborg Trailer became a useful exploration of where AI can already be effective in film-oriented production, and where traditional workflows are still necessary.