Cyberpunk City AI

I’ve always enjoyed creating images for Cyberpunk City, capturing its different districts and the many aspects of everyday life within the world. I learned a lot about both the possibilities and limitations of AI image generation, especially after I started experimenting with image generation in ChatGPT in 2024. At first, I was simply creating individual scenes and exploring what Cyberpunk City could look like, but eventually I wanted to tell stories within that world.

One of the biggest challenges has been maintaining consistency. AI has a tendency to recreate a scene rather than remember it exactly, which means details can shift between images. When I started creating stories with recurring characters, dialogue, and connected scenes, I quickly realized how difficult it was to keep characters looking the same. Their appearance could change dramatically as the story progressed, even when I was trying to recreate the same person.

Eventually, I became frustrated with the limitations of the technology and gave up on the idea for a while. But I began noticing other creators producing surprisingly consistent characters with AI, and that motivated me to learn more about the techniques they were using and experiment with my own process.

When I restarted Cyberpunk City AI, I began with four main characters. I created reference images and character sheets for each of them to help establish their appearance and keep them as consistent as possible. This has helped a lot, but current AI technology, including ChatGPT, still requires multiple prompts and adjustments to maintain consistency from panel to panel. Sometimes a character’s face, clothing, cybernetics, or even small features need to be corrected several times before the image feels right.

Keeping scenes consistent is even more challenging. I use reference images whenever possible, especially for important locations, but maintaining small details between comic panels can be difficult. AI may move objects, change the background, alter lighting, or subtly redesign parts of a location from one image to the next. Sometimes these changes are minor, but when you’re telling a continuous story, they can become noticeable.

I occasionally use Adobe Firefly to spot-remove errors or make small corrections after generating an image, although that process is imperfect as well. In the end, I’ve learned that I have to let go of perfection and accept some inconsistencies. Otherwise, I could spend an endless amount of time trying to fix every small detail instead of actually telling the story.

Currently, creating a single three-panel comic usually takes me around two hours. That includes writing the dialogue, developing the story, generating the images, and then spending a significant amount of time using additional prompts to modify and refine the images until they match what I have in mind.

In the end, Cyberpunk City is just a hobby for me, and it doesn’t have to be perfect. I’ve found a lot of joy in creating a world and slowly bringing it to life. Even if there aren’t many people viewing the stories, I enjoy exploring this cyberpunk world that mostly exists in my imagination and sharing pieces of it as I continue to build it.