How to Write AI Image Prompts: Complete Beginner's Guide

Most people write AI image prompts the wrong way. They type 'Ghibli style portrait' and get a generic, inconsistent result — then blame the AI. The truth is that AI image models are extremely literal: they generate exactly what you describe, nothing more. The difference between a mediocre result and a stunning one is almost entirely in how you write the prompt. This guide explains the full anatomy of a high-quality prompt, step by step.

The Five Core Components of Any Good Prompt

Every strong AI image prompt contains five elements: style, subject, lighting, mood, and camera/composition. Skipping any of these leaves the AI to guess — and it will guess generically. Here's what each element does:

• Style — The visual aesthetic: Ghibli, cinematic, oil painting, watercolor, etc. This is the 'what does it look like' element
• Subject — What or who is in the image: 'portrait of a woman', 'a golden retriever', 'product shot of a perfume bottle'
• Lighting — The light source and quality: golden hour, soft studio light, dramatic chiaroscuro, neon glow
• Mood — The emotional tone: serene, dramatic, nostalgic, playful, melancholic
• Camera — Framing and lens: close-up, wide shot, 85mm lens, shallow depth of field, bird's eye view

Style: More Than Just a Name

Writing just 'cinematic style' tells the AI almost nothing. Cinema spans 100 years and hundreds of visual aesthetics. The more you specify, the more accurate the result. Compare these two prompts:

Bad:  cinematic portrait
Good: cinematic film still, 1970s New Hollywood aesthetic, anamorphic lens flare, warm tungsten light, shallow depth of field, film grain texture, Kodak color grading

When specifying style, mention: the medium (film, painting, animation, photography), the era or movement if relevant, specific textural qualities (grain, brushstrokes, ink lines), and any color profile or palette.

Lighting: The Most Underrated Element

Lighting changes the emotional register of an image more than almost any other element. The same portrait taken in golden hour light looks romantic and warm. Under harsh overhead light it looks clinical. Under dramatic side light it looks intense and powerful. Be explicit.

• Golden hour — warm, directional, long shadows, romantic or nostalgic mood
• Soft diffused — overcast or studio softbox, flattering, even, gentle
• Dramatic chiaroscuro — strong contrast, one-sided light, intense and cinematic
• Backlit — rim light, silhouette or halo effect, dreamlike
• Neon — urban night, coloured light sources, cyberpunk or moody
• Magic hour / twilight — deep blues, fading light, melancholic or peaceful
• Hard overhead — harsh shadows, raw, documentary or tense

Mood: Use Emotion Words, Not Just Adjectives

Mood words tell the AI how to weight all the other choices. 'Dramatic' pushes toward high contrast, strong shadows, intense expressions. 'Nostalgic' pulls toward warm tones, soft edges, vintage textures. 'Playful' brightens the palette and loosens the composition.

The most effective mood descriptors are specific emotional states rather than generic adjectives. Instead of 'beautiful', write 'serene and melancholic'. Instead of 'cool', write 'ominous and tense'. These give the model emotional direction.

Camera: Think Like a Photographer

Camera descriptors dramatically affect composition and depth. Most people skip these — which is why their results look 'AI-generated' rather than photographically composed. Add at least a focal length or framing description.

• 35mm lens — natural field of view, slight environmental context, documentary feel
• 50mm lens — closest to human eye, neutral, balanced
• 85mm portrait lens — slight compression, flattering for faces, background blur
• 135mm telephoto — strong background compression, intimate feel, isolated subject
• Wide angle (24mm) — dramatic perspective, spatial context, environmental
• Macro — extreme close-up, texture detail, abstract quality
• Bird's eye view — overhead perspective, pattern and geometry emphasis
• Low angle — subject appears powerful, monumental

Model-Specific Formatting

Different AI tools process prompts differently. What works in ChatGPT will give suboptimal results in Midjourney if you don't adjust the format.

• ChatGPT 4o — use full sentences and structured paragraphs. It responds to conversational instructions. Say 'Preserve the subject's facial identity' explicitly for portraits
• Midjourney — use comma-separated keywords. Add --ar for aspect ratio (e.g. --ar 4:5), --style raw for less processing, --v 7 for latest model
• Gemini — similar to ChatGPT, responds to natural language. Specify 'high quality', 'detailed' explicitly
• FLUX — most literal of all models. Describe exactly what you see in the image, element by element. Very detailed prompts work best

Prompt Length: How Long Is Too Long?

A common misconception is that longer prompts are always better. The truth is that relevance matters more than length. A 200-word prompt stuffed with contradictory style references will perform worse than a focused 60-word prompt that covers all five core components clearly.

A good target is 40–80 words for most use cases. This is enough to cover style, subject, lighting, mood, and camera without becoming internally contradictory. The exception is Midjourney, where compact keyword-style prompts (20–40 words) often outperform long sentences.

Common Prompt Mistakes

• Conflicting styles — 'Ghibli cyberpunk oil painting' gives a confused muddy result. Pick one dominant style
• Vague subjects — 'a person' is too vague. Specify: 'a young woman in her 20s, dark curly hair'
• Missing lighting — without lighting info, the AI defaults to flat, unmotivated light
• Forgetting to preserve faces — for ChatGPT transformations, always add 'preserve facial identity'
• Over-engineering — adding 20+ style modifiers cancels each other out
• Copy-pasting without context — Midjourney prompts don't work the same in ChatGPT and vice versa

Prompt writing is a skill that improves with practice. The fastest way to improve is to take a result you don't like and ask yourself: which of the five elements (style, subject, lighting, mood, camera) was I vague about? Add specificity there, and regenerate. You'll see improvement immediately.