Writing Prompts That Actually Work

The gap between a mediocre AI image and an impressive one often comes down to the prompt. This lesson covers the structure and vocabulary that reliably produces better results.

The Anatomy of an Image Prompt

Effective image prompts typically specify:

1. 1.Subject — what is in the image and what it is doing
2. 2.Style — the visual aesthetic, art movement, or medium
3. 3.Composition — camera angle, framing, what is in the foreground/background
4. 4.Lighting — the quality and direction of light
5. 5.Mood — the emotional tone of the image
6. 6.Technical parameters — quality modifiers, aspect ratio, specific tool settings

Not every prompt needs all six. But adding underspecified elements usually improves results.

Subject: Be Specific

Vague: "a woman" Better: "a woman in her 40s, confident expression, wearing a blazer, looking off-camera" Even better: "a woman in her 40s, confident expression, wearing a dark navy blazer, looking slightly off-camera to the left, shallow depth of field background"

Specificity gives the model less room to guess, and its guesses are often not what you want.

Style Vocabulary Worth Knowing

• Art movements and styles:
• Hyperrealistic photography, photorealistic, cinematic
• Impressionist, oil painting, watercolour, gouache
• Art nouveau, art deco, bauhaus
• Concept art, digital painting, matte painting
• Flat design, vector art, minimalist illustration

• Photography terms that work well:
• Shot on 35mm film / shot on a Sony A7R / Leica photograph
• Golden hour lighting, blue hour, overcast diffused light
• Wide angle, telephoto compression, macro lens
• Bokeh background, long exposure, HDR

• Mood and atmosphere:
• Ethereal, moody, atmospheric, dramatic, serene, tense

The Style Reference Technique

Instead of describing a style from scratch, reference artists or works whose aesthetic you want:

• "in the style of Edward Hopper"
• "Studio Ghibli aesthetic"
• "Wes Anderson colour palette"
• "Shot in the style of Annie Leibovitz"

Be aware: many tools are moving away from direct artist name references for copyright reasons. Describing the characteristics of a style is often more reliable than naming the artist.

Negative Prompts: Telling the Model What to Avoid

• Most tools allow you to specify what you do not want in the image. This is particularly useful for:
• "no text, no watermark"
• "no extra fingers, no distorted hands"
• "no busy background"
• "no oversaturated colours"

In Midjourney, use the --no parameter. In Stable Diffusion and DALL-E, there is usually a dedicated negative prompt field.

Iteration: How to Refine Results

Rarely does the first prompt produce the perfect image. The workflow is iterative:

1. 1.Run with your initial prompt
2. 2.Identify what is right and what is wrong
3. 3.Adjust the prompt to address the issues
4. 4.Run again

• Common adjustments:
• If the composition is wrong: add "close-up", "wide shot", "bird's eye view", "portrait orientation"
• If the style is off: add more specific style references
• If the lighting is flat: add "dramatic lighting", "golden hour", "soft diffused light"
• If it looks too AI-generated: add "photorealistic", "photograph", "realistic photograph"