The author proposes a method for maintaining a consistent visual style across a series of AI-generated images by extracting a 'genome'—a detailed description of the visual language. This process involves breaking down a reference image into specific features such as lighting, texture, and color, while excluding subject-specific details.
The methodology includes identifying the 'core' elements of a style that must remain constant and 'variables' that can be changed. By creating a master prompt with placeholders for subjects and actions, users can generate diverse content while preserving the original aesthetic. The process also incorporates a negative prompt to prevent the model from reverting to default generation habits.
Once the master prompt is tested across various objects, it can be saved as a 'skill'—a reusable instruction set. This approach aims to streamline the creation of image series by providing a repeatable process, though the author notes that results may vary between different AI models and that some complex visual nuances remain difficult to describe in text.
Separately, the author discusses the 'Cost per Task' metric for AI models. Comparing models like GPT-5.6 Luna (max) and GPT-6 Astra (max), the author notes that the former costs $0.18 per task compared to $3.26 for the latter. The author suggests that using more efficient models for well-defined tasks can significantly reduce operational costs.
