Start with the subject, then add the setting and any details that matter, such as lighting, colours, or medium. For example: “a red fox asleep on a mossy log, morning light, 35mm photograph”.The model uses an encoded representation of your words during generation. It can miss details or combine them in unexpected ways. Wording, order, and supported emphasis syntax can affect the result, but there is no universal rule that the first words always matter most.Try a clear description first. Add or change a phrase when you can identify what is missing. Models respond differently to the same prompt, so keep the model fixed when comparing wording.