Model-Aware Prompt Layer: Tailoring One Image Idea for Seven Models

Prompts that work well in one image model often underperform in another, since each model responds best to different emphasis such as composition, text rules, or motion language. The author built a small model-strategy layer, a lookup table that sits between the user's request and the instructions sent to the model. Each model gets a single strategy sentence while a shared prompt structure keeps the rest consistent, so adding a new model requires only one new line. The approach includes a deterministic local fallback that still produces a usable prompt if the AI call fails. The write-up walks through the implementation with code from the project.