How to Write AI Image Prompts That Actually Work

Start with the subject, not the style

The single most common mistake in AI image prompting is leading with style words. "Beautiful, amazing, ultra detailed" tells the model almost nothing. Start with a concrete subject and a clear action: a fisherman mending nets, a ceramic mug on a stone pedestal, a dragon over a valley.

Add setting, composition, and light

Once the subject is fixed, layer in three things. First the setting: where is the subject and what surrounds it. Second the composition: close-up, wide establishing shot, centered with negative space. Third the lighting: golden hour backlight, soft studio key light, neon reflections on wet pavement. Lighting does more for perceived quality than any quality keyword.

Finish with style and technical cues

Only now add the rendering style: photorealistic, watercolor, fine line tattoo, 85mm lens at f/1.8. Then use a negative prompt for what you do not want: blurry, watermark, extra fingers. This order — subject, setting, composition, light, style — mirrors how photographers and illustrators actually plan an image, and models respond to it far better than keyword soup.

Use structure when it helps

If you think in data, write your prompt as JSON with fields like subject, setting, lighting, style, and negative_prompt. Our generator reads both plain text and structured JSON, so use whichever keeps your ideas clearest.