Anatomy of a Good Image Prompt
Most beginners type a four-word prompt, see a mediocre result, and conclude AI image generation is overrated. The actual problem is the prompt. A good image prompt is a sentence or two that tells the model six specific things — and once you learn the formula, your images jump from "okay" to "share this" overnight.
What You'll Learn
- The six ingredients every strong image prompt contains
- A copy-paste template you can use immediately
- Common beginner mistakes and how to avoid them
- A drill for practicing prompt-writing in 10 minutes
The Six Ingredients of a Great Prompt
Think of your prompt as a stage direction. The model is the cinematographer. Tell it:
1. Subject — what or who is in the image "A red panda" / "A cyberpunk samurai" / "A bowl of ramen"
2. Action or pose — what the subject is doing "...sitting cross-legged on a tatami mat..."
3. Setting and environment — where it is "...inside a quiet Kyoto teahouse, sliding paper doors in the background..."
4. Style or medium — how it should look "...studio Ghibli watercolor illustration..."
5. Lighting and mood — the emotional feel "...soft afternoon light filtering through paper screens, peaceful, warm..."
6. Technical specs — camera angle, framing, aspect ratio "...wide shot, eye-level, 16:9 cinematic composition..."
Stitch them together and you get:
A red panda sitting cross-legged on a tatami mat inside a quiet
Kyoto teahouse, sliding paper doors in the background, Studio
Ghibli watercolor illustration, soft afternoon light filtering
through paper screens, peaceful and warm mood, wide shot at
eye level, 16:9 cinematic composition.
That single prompt will produce an image that genuinely looks intentional. Compare it to "red panda in a teahouse" and you'll see the difference instantly.
The Copy-Paste Template
Save this somewhere you can reach quickly. Fill in the blanks:
A [subject] [doing what] in [setting], [style or medium],
[lighting], [mood], [camera framing], [aspect ratio].
Examples filled in:
A young woman with short curly hair laughing at her laptop in
a cozy university library, oil painting, golden hour light,
warm and hopeful mood, medium close-up, 3:2 portrait composition.
A futuristic electric scooter parked on a wet Tokyo street at
night, photorealistic, neon reflections in puddles, moody and
atmospheric, low-angle shot, 16:9 widescreen.
The Five Most Common Beginner Mistakes
Mistake 1: Vague subject Bad: "a dog" Better: "a golden retriever puppy with floppy ears, looking up curiously"
Mistake 2: No style direction Bad: "a city at night" Better: "a city at night, cyberpunk illustration style, neon palette"
Mistake 3: Conflicting instructions Bad: "minimalist but extremely detailed and cluttered" Better: pick one and commit — the model gets confused by contradictions.
Mistake 4: Negative prompts as part of the description Bad: "a beach with no people and no sand and no sun" The model often ignores negations and might produce exactly what you said you didn't want. In tools that support it (Midjourney, Stable Diffusion, Leonardo), use a separate negative prompt field. In ChatGPT and Gemini, just describe what you DO want.
Mistake 5: Trying to dictate rather than describe The model doesn't think in instructions like "make sure to include exactly seven trees." Instead describe a scene that naturally contains seven trees, or use a smaller specific number like "three" — the model handles small counts much better than large ones.
Style Keywords That Actually Work
These phrases produce reliably distinct looks. Mix and match:
- Photorealistic / 8K photo / DSLR / shot on Hasselblad / shot on iPhone 16 Pro
- Watercolor / oil painting / gouache / pencil sketch / ink illustration
- Studio Ghibli / Pixar 3D / Disney 2D animation / anime / manga
- Cyberpunk / steampunk / vaporwave / dark academia / cottagecore
- Minimalist line art / flat vector / isometric illustration / 3D render
- Polaroid / 35mm film / black and white / sepia / infrared
Lighting keywords:
- Golden hour / blue hour / harsh midday sun / moonlight / studio softbox
- Backlit / rim light / volumetric light / god rays / chiaroscuro
- Neon / candlelight / fluorescent / window light / firelight
Drill: The 10-Minute Prompt Workout
Spend 10 minutes today on this exercise, and your prompts will permanently improve.
- Open ChatGPT or Gemini.
- Take this base prompt: "A coffee shop in autumn."
- Generate one image with this short prompt. Save it as v1.
- Now expand it using the template — subject, action, setting, style, lighting, mood, framing, aspect ratio. Generate the expanded version. Save it as v2.
- Now change ONLY the style ("...as a watercolor painting" → "...as a cyberpunk neon scene"). Save as v3.
- Now change ONLY the lighting ("...in golden hour" → "...in a thunderstorm at night"). Save as v4.
Compare v1 to v4. You just learned the four levers — subject, expansion, style, lighting — that drive 80% of an image's quality.
A Pro Tip: Let AI Write Your Prompts
When you can't think of style words, paste this into ChatGPT or Claude:
You are an expert AI image prompt engineer. Write me three
detailed image prompts for: [your idea]. For each, vary the
style and mood. Use the format: subject, action, setting,
style, lighting, mood, framing, aspect ratio.
Then paste the prompts into your image tool. This trick alone is worth the price of admission — and the price was zero.
Key Takeaways
- A great prompt names six things: subject, action, setting, style, lighting, mood, framing, aspect ratio
- Specific is better than vague; describe what you want, not what you don't
- Style and lighting keywords drive most of the visual difference between mediocre and great results
- Use ChatGPT or Claude to write prompts for you when stuck — meta-prompting is a free superpower