Game Cover Generator
An interactive, Smash Melee inspired interface that uses pre generated AI covers and a procedural title system to explore hybrid human and machine authorship.
Reflection
For this project I wanted to build something that lives in between a GameCube start screen and a modern AI tool. The Game Cover Generator lets you move through a menu that feels like Smash Melee, choose a genre, a visual style, a color palette, and a decade, and then watch a cover and title appear. The browser is not actually talking to Midjourney in real time. Instead, I treated the whole interface as the main artwork and used a small set of pre generated covers as the visible trace of my collaboration with AI.
I was interested in this idea because game covers have always felt like promises to me. Before you ever play a game, the box art and logo tell you how serious it is, what kind of world you are walking into, and what kind of player it expects you to be. At the same time, current AI image tools can spit out endless fake covers in a few seconds. I wanted to hold those two feelings at once. On one side there is nostalgia and slow, intentional design. On the other side there is speed, randomness, and models that remix millions of images that I will never see.
The workflow for this project was hybrid from the very beginning. I used ChatGPT as a coding partner to help sketch out the basic structure of the page. Together we set up the different screens, the state that remembers each choice, and the JavaScript that builds a title from different word pools. After that I kept pushing and changing what it gave me. I rewrote button text, adjusted spacing, rebalanced the glow and shadows, and stepped in whenever the layout felt too busy or too flat. The code that is running now is not a simple copy from AI. It is the result of going back and forth, and making design calls that match my own taste and memories of older game menus.
On the visual side, Midjourney acted like an off screen art team. I turned the UI options into very specific prompts. For example, I generated images for things like a retro RPG pixel castle, a modern painterly RPG party, a modern pastel sci fi astronaut, a glitchy horror face, and a red house in the woods. From each batch I picked only the covers that felt like they could sit on a real shelf. The generator page does not invent new images each time. It simply maps your choices to this small curated pool. That choice matters, because it keeps the project honest and finite instead of pretending to be an endless AI machine.
While I was building this, I realized that the interesting part was not the illusion of generation but the way the interface controls pacing. At first I imagined everything on a single page, with one form and one button. Once I leaned into separate screens, the project clicked into place. The Press Start title screen, the genre menu, the style menu, the palette menu, the decade menu, and finally the reveal screen all shape your expectations step by step. That structure reminded me that interfaces are not neutral containers. They decide what feels like choice, what feels like commitment, and how much control you think you have over what the machine will do.
This connects directly to our class discussions about hybrid authorship and entangled aesthetics. None of the final covers are purely mine, but they are also not just the model’s work. I did not paint every pixel, but I did choose the categories, write and refine the prompts, sort through outputs, and decide which images earned a place in the set. The generator then combines those images with a dynamic title system that is driven by the genre, style, palette, and decade you select. In that way, the project behaves like a small authorship system. There is human curation, model output, and procedural remix all working together.
The menu flow is also a quiet argument about classification. You never type free text into this generator. You move through fixed options like RPG, Horror, Pixel Art, Realistic 3D, Neon, Earthy, 1990s, or 2020s. By the time you reach the end, your cover feels personal, but it is actually the result of a very specific taxonomy that I designed and a bit of randomness inside each category. That mirrors how large models work. They feel open and flexible, but they always depend on someone’s labels, training data, and constraints behind the scenes.
Working with AI for this project also brought up ethical questions that shaped my choices. Midjourney’s training data is not fully transparent, and it can easily imitate the style of artists who never consented to that use. I did not want to present these images as if I had painted them by hand. The interface hints that the images are AI assisted, and this documentation says that out loud. I also kept the image pool small on purpose. When you spend enough time with the generator, you start to recognize covers and notice the limits of the system. That felt more honest than a fake sense of infinite variety.
On a personal level, this project changed how I think about my own creativity with AI tools. The most satisfying moments were not when the model gave me something perfect on the first try. They were the moments where something was broken or ugly, and I had to fix it. That could mean debugging an image path that did not load, toning down an over the top glow, or throwing away a cover that looked cool but did not fit the rest of the set. Those decisions came from my taste, my sense of game history, and my memories of specific menus and box art. AI helped me move faster and explore more variations, but it did not remove the need for judgment or direction.
In the end, Game Cover Generator is less about showing off what AI can do visually and more about exposing the structure around it. The menus, the choice states, and the mapping from options to images all make visible the work that usually happens off screen. Someone has to decide what counts as a genre, what kind of art fits a label like Pastel, and which images get to represent a decade. This project reminded me that when I work with AI, I am not just writing prompts. I am also building the frame that tells people how to read whatever the model gives back.
AI Tools Used
- ChatGPT (OpenAI): I used this to brainstorm the concept, scaffold the HTML, CSS, and JavaScript, debug layout problems, and draft this reflection text, which I then edited in my own voice.
- Midjourney: I used this to generate the video game cover images. The prompts matched the genre, style, color palette, and decade options that appear in the interface.