Project Overview
The goal of this project was to test the limits of our V4 model's capability to generate complex, multi-layered urban environments with specific lighting conditions. We aimed to create a cohesive series of images depicting a cyberpunk city that feels lived-in and historically layered, rather than just a collection of neon lights.
The Challenge
Generating consistent architectural styles across multiple images is notoriously difficult with diffusion models. Often, the AI will halluncinate different eras or physics-defying structures. We needed to maintain a consistent "tech-noir" aesthetic while varying the camera angles and subject matter.
The Solution
We utilized a combination of specific "negative prompts" to filter out unwanted architectural styles and "seed locking" to maintain consistent color palettes. TeraAI's new "Style Reference" feature allowed us to feed the first successful generation back into the model as a constraint for subsequent images.
