The Seven Ingredients of a Great AI Image Prompt

Type "a cat" into an image generator and you get a cat — the statistical average of every cat picture on the internet, in the model's favourite lighting, from its favourite angle. Technically what you asked for. Nothing like what you imagined.

The gap between "a cool picture" and "the picture in your head" is closed by seven decisions. Make them explicitly and any decent generator will follow you. Skip them and the model decides for you — always in the direction of generic.

The Seven Ingredients

1. Subject — who or what, specifically

Not "a cat": "an elderly ginger street cat with a torn ear". Specificity in the subject is the difference between the average of all cats and this cat. Give it age, texture, character, one imperfection.

2. Setting — where, and when

"On a rain-wet Colombo street at dusk, tuk-tuks blurred in the background." Setting anchors every other choice the model makes — colours, reflections, atmosphere. An unspecified setting is why so many AI images float in beige nowhere.

3. Style — what kind of image this is

Photograph? Watercolour? Flat vector illustration? 35mm film still? Pixel art? This single ingredient changes the output more than any other — and leaving it out hands the model its default: that glossy, over-polished "AI look".

4. Light — the professional's secret

"Soft golden-hour side light." "Harsh single fluorescent tube." "Candlelit, deep shadows." Photographers obsess over light because light is mood made visible; the moment you start specifying it, your images stop looking amateur. Most beginners never mention light at all — see Why Your AI Images Look Generic for what that costs.

5. Mood — the emotional temperature

Melancholy, jubilant, tense, serene, nostalgic. Models respond strongly to emotional vocabulary — it shifts palette, posture, weather, framing, everything at once. One mood word is the cheapest big lever in prompting.

6. Shot — the camera's position

Close-up, wide establishing shot, overhead, low angle looking up, over-the-shoulder. Borrowed from cinema, and it works: "low-angle wide shot" produces drama that "picture of" never will. If you've ever wondered why your images all feel identically framed — you never chose a frame.

7. Detail — the finishing pass

The specifics that sell reality or style: "steam rising from the teacup", "grain and slight motion blur", "hand-painted texture". One or two well-chosen details; a list of twenty just muddies the signal.

Watch It Work

"A cat" → generic. Now stack the ingredients: "An elderly ginger street cat with a torn ear (subject), on a rain-wet Colombo street at dusk (setting), shot like a 35mm film photograph (style), lit by shop-window neon reflecting off the wet road (light), melancholy and quiet (mood), low-angle close-up (shot), whiskers beaded with drizzle (detail)."

Same generator. Different universe.

Learn It by Doing, Not Reading

Here's the thing about these seven: reading them nods your head, but seeing what each one changes is what installs the skill. That's exactly how Cocoon's free Creative Lab teaches it — you build prompts ingredient by ingredient and generate real images live as you learn, watching the output transform as each ingredient lands. Pick your level, no signup, genuinely fun.

The same ingredient thinking extends to AI video — with a few extra rules covered in AI Video Prompting for Beginners. And the general theory of why structured prompts beat vibes, for text as much as images, is in The Anatomy of a Perfect Prompt.

Enter the Creative Lab — and stop letting the model make your creative decisions.