Image Prompt Rules

Different image generators handle prompts differently. The rules below start with what holds for every model, then cover each family's weighting syntax.

Universal Rules (every model)

• Word order: earlier words carry more weight — lead with your main subject.
• Quantity control: keep main subjects to two or three per image; more tend to interfere with each other.
• Length: concise prompts (within ~380 characters) reproduce more reliably and are easier to iterate on.

Stable Diffusion Family (WebUI, ComfyUI, etc.)

• Parenthesis weighting: (word) adds light emphasis and stacks (e.g. ((word))); (masterpiece:1.5) sets an exact multiplier.
• Blend / switch: square-bracket syntax. [blue hair|white hair] alternates two elements step by step for a blended look; [blue hair:white hair:0.3] switches from blue to white hair at 30% progress.
• Negative prompts: put unwanted elements in the dedicated negative-prompt box (e.g. blurry, bad anatomy, extra fingers). IMGPrompt's Negative button copies a ready-made set.

Midjourney

• Weights: phrase::number, e.g. sunset::2 desert::1 makes the sunset twice as important as the desert.
• Excluding elements: use the --no parameter, e.g. --no tree for a scene without trees.

Natural-Language Models (FLUX, GPT-Image, Veo, Kling…)

No special symbols needed — describe the scene in full, specific sentences, put what matters first, and state exclusions in plain words (e.g. "no trees in the scene"). Tags for camera, lighting and art style from the library still slot right in.