AI Tools & Software

How to Write Better Image Prompts: Subject, Style, Lighting and Composition

Abstract blue image graphic with the label Write Better Image Prompts

Most disappointing AI images come from vague prompts. "A coffee shop" produces a generic stock-photo look; "a small corner coffee shop at 7 am, warm window light, steam rising from a cup on a wooden counter, shallow depth of field, 35mm photo" produces something you might actually use. Learning to write better image prompts is the fastest way to improve results in any AI image generator. This guide gives you a simple structure, vocabulary for style and lighting, examples and fixes for common problems.

The six-part prompt structure

Good image prompts usually cover six things, roughly in this order:

  1. Subject: who or what, doing what.
  2. Setting: where and when.
  3. Style: photo, illustration, 3D, watercolor, flat vector.
  4. Lighting: soft daylight, golden hour, studio, neon.
  5. Camera and composition: close-up, wide shot, overhead, rule of thirds.
  6. Details and mood: colors, textures, emotion, things to avoid.

You will not need all six every time, but checking against the list catches what is missing.

Write better image prompts: subject and setting

Be concrete:

  • Weak: "a woman working".
  • Better: "a woman in her 40s reviewing invoices at a kitchen table, laptop open, morning light".

Name the action, the key objects and the environment. Avoid real people's names and identifiable private individuals.

Style vocabulary

  • Photography: "35mm photo", "product photo on white background", "documentary style".
  • Illustration: "flat vector illustration", "children's book illustration", "line art".
  • Painting: "watercolor", "oil painting with visible brushstrokes".
  • 3D: "soft 3D render", "isometric 3D".
  • Graphic design: "minimal poster", "infographic style".

Describe styles rather than naming living artists.

Lighting vocabulary

  • Soft natural light for calm, realistic scenes.
  • Golden hour for warm, flattering light.
  • Studio lighting for products.
  • Backlit or rim light for drama.
  • Overcast for even, shadow-free images.

Lighting changes mood more than almost anything else.

Camera and composition

  • Shot size: extreme close-up, close-up, medium, wide.
  • Angle: eye level, overhead, low angle.
  • Lens feel: wide angle, telephoto, shallow depth of field.
  • Composition: centered, rule of thirds, negative space for text.

If you need space for a headline, ask for "negative space on the left for text".

Examples

Product photo: "Ceramic coffee mug in matte sage green on a light oak table, soft window light from the left, shallow depth of field, minimal styling, product photography, negative space on the right."

Blog illustration: "Flat vector illustration of a small business owner at a desk surrounded by floating icons for email, calendar and invoices, soft blue and white palette, clean lines."

Thumbnail background: "Bright, high-contrast scene of a cluttered desk with papers flying, playful mood, saturated colors, wide shot, space on the right for a face and text."

Fixing common problems

  • Too generic: add specific details about subject, setting and style.
  • Wrong mood: change lighting and color words first.
  • Cluttered: ask for "minimal", "clean background" or "simple composition".
  • Weird hands or faces: choose crops that hide hands, or regenerate; use inpainting to fix small areas.
  • Bad text: shorten it, put it in quotes, or add text later in a design tool.
  • Inconsistent series: reuse the same style and lighting phrases.

Using references

Many tools accept reference images for style or composition. Use your own photos or images you have rights to, and keep a library of references that match your brand.

Iterating efficiently

  1. Start with a short prompt and generate a few variations.
  2. Change one element at a time: lighting, then style, then composition.
  3. Save prompts that work in a document with the resulting image.
  4. Build reusable templates for recurring needs, such as blog headers or product shots.

Tool-specific notes

Different generators respond differently to the same prompt. Some excel at photorealism, some at text, some at illustration. For comparisons, see our Ideogram vs Midjourney guide, and for product shots specifically, our guide to AI product photos. For thumbnails, see our AI thumbnail maker guide.

Prompting for business use cases

Blog headers: ask for a consistent style across posts, such as "flat vector illustration, soft blue palette, simple shapes, negative space at the top for a title". Reusing the style phrase keeps your blog visually consistent.

Social posts: specify the platform's shape and leave space for text where you plan to add it in a design tool.

Ads: avoid realistic people unless you have rights and disclosure in place; use product-focused scenes or illustrations instead.

Presentations: ask for simple, uncluttered backgrounds that do not compete with slide text.

Building a prompt library

Keep a document with three columns: the prompt, the image it produced, and notes on what worked. Over a few weeks you will build reliable recipes for each type of image you need, and anyone on your team can reuse them. That library becomes more valuable than any single image.

Negative prompts and exclusions

Some tools let you list what to avoid. Even when they do not, adding "avoid text, logos and watermarks" or "no people" to the end of a prompt often helps. Keep exclusions short and specific.

Editing instead of regenerating

When an image is nearly right, edit it rather than starting over. Inpainting can fix a hand or replace an object; cropping can remove a distracting edge; a design tool can add clean text. Small edits are faster and keep the parts of the image you already like.

Ethics and rights

  • Avoid real people without permission.
  • Avoid trademarks and copyrighted characters in commercial work.
  • Describe styles rather than imitating living artists.
  • Check each tool's commercial terms before using images in paid work.
  • Label realistic AI images where platforms require it.

A prompt template to copy

"[Subject doing action], [setting and time], [style], [lighting], [shot size and angle], [composition notes], [color palette and mood], [avoid: ...]."

Example: "A florist arranging tulips, small shop interior in the morning, documentary photo style, soft window light, medium shot at eye level, subject on the left with space on the right, fresh pastel colors, calm mood, avoid text and logos."

Related reading

For a wider look at AI video and channel workflows, see our guide to a faceless YouTube channel with AI.

FAQ

What makes an image prompt good?

Specific details about subject, setting, style, lighting, camera angle and composition, written in clear language, plus anything to avoid.

Do longer prompts work better?

Not always. Specific, well-ordered details matter more than length. Remove words that do not change the image.

How do I get text in an image right?

Keep text short, put it in quotation marks, and choose a model known for good text rendering. Check spelling in every output.

Can I use artist names in prompts?

Describe the style instead, such as 'loose watercolor with soft edges'. Imitating living artists raises ethical and sometimes legal concerns, especially for commercial use.

How do I keep a consistent style across images?

Reuse the same style phrases, use reference images where the tool supports them, and keep a document of prompts that worked.

Conclusion

To write better image prompts, describe the subject, setting, style, lighting, camera and composition, change one element at a time, and save the prompts that work. With a simple structure and a growing prompt library, AI images move from generic to genuinely useful.

Sources and methodology

Practical guide based on how current AI image generators respond to prompts and common photography and design vocabulary. No statistics are claimed.