How to Write AI Image Prompts

Image prompt quality is the difference between “mediocre AI slop” and “people genuinely can’t tell.” Here’s how to write prompts that work across Midjourney, DALL-E 4, FLUX, Imagen, and Stable Diffusion 3.5 in 2026.

The 5-part structure

Stack style + subject + lighting + camera + extras + flags. A simple working prompt:

cinematic still, anamorphic lens, color graded,
a labrador wearing a tiny astronaut helmet,
golden hour warm side light,
85mm portrait lens shallow depth of field,
dust particles in the air, eye contact,
--ar 16:9

What each piece does

• Style: “photo”, “cinematic still”, “3D render”, “anime” sets the broad aesthetic.
• Subject: describe what the picture IS in plain language.
• Lighting: golden hour, studio softbox, neon, moody low-key — the single biggest quality lever.
• Camera: 24mm wide / 85mm portrait / macro / fisheye dictates composition feel.
• Extras: texture, atmosphere, mood notes (e.g., “dust particles”, “eye contact”).
• Flags: Midjourney --ar, --no, --style raw; SD’s separate negative prompt.

Common mistakes

• Asking for “a photo of a man” with no other detail. Models default to generic — you get slop.
• Stacking too many adjectives. After ~30-40 tokens, models trail off on detail adherence.
• Putting subject last. Most models weight earlier tokens more heavily.
• Using artist names without checking copyright comfort — commercial use risk.

Build prompts faster with the AI image prompt helper. For full tool comparison see best AI for image generation.