Prompt engineering started as a name for "what gets the model to do what I want" and has evolved into a real skill. For interior design, the prompt is what tells the diffusion model what style, materials, lighting, and mood to aim for. Small wording differences materially change the output.
A few patterns that work for design prompts: name specific materials rather than abstract styles ("smoked white oak floorboards" beats "wood floor"); specify lighting source and time of day ("late-afternoon sun through west-facing windows"); name furniture archetypes rather than brands ("low-slung sectional sofa with rounded arms" beats "Italian sofa"); and constrain composition with a camera direction ("wide-angle, 35mm, eye level").
Good AI interior design tools hide most of this behind preset buttons — you click "Scandinavian" and the underlying prompt is professionally engineered. But if you're using a raw model (Stable Diffusion, Midjourney) you write the prompt yourself, and the difference between a poor and a strong prompt is the difference between stock-photo output and something that actually feels designed.