How to Write AI Image Prompts (Formula, Examples and Templates)

Quick answer: A good AI image prompt describes the subject, what it is doing, the setting, the style or medium, the lighting, the composition or camera angle, the colors and mood, and the format you need, with the most important details first. Say what you want rather than what you don’t, be specific about materials and light, then refine the result with short follow-up edits.

What is an AI image prompt?

An AI image prompt is the text description you give an image generator such as ChatGPT’s image tool, Google Gemini, Midjourney, Adobe Firefly, Stable Diffusion, Flux or Canva’s AI features. The model turns your words into a picture, so everything you leave out is filled in with the model’s defaults, which is why vague prompts produce generic images.

How image generators read your prompt

• Early words carry more weight: many generators prioritize what appears first, so lead with the subject.
• Concrete beats abstract: “soft window light from the left” is actionable; “nice lighting” is not.
• Conflicts cause compromises: “minimalist” and “busy, detailed scene” in the same prompt produce muddled results.
• Negation is unreliable: writing “no cars” can make cars more likely in some tools; describe an empty street instead, or use a dedicated negative prompt field if your tool has one.

The image prompt formula

Part	Question it answers	Example
Subject	Who or what?	A ceramic coffee mug
Action or state	What is happening?	with steam rising
Setting	Where and when?	on a wooden café table, early morning
Style or medium	What kind of image?	editorial product photography
Lighting	How is it lit?	soft window light from the left
Composition	How is it framed?	close-up, shallow depth of field, subject off-center
Color and mood	How should it feel?	warm neutrals with a teal accent, calm
Format	Size and use?	16:9 blog header, empty space on the right

Put together: “A ceramic coffee mug with steam rising, on a wooden café table in the early morning. Editorial product photography, soft window light from the left, close-up with shallow depth of field and the mug off-center. Warm neutral tones with a teal accent, calm mood. 16:9 blog header with empty space on the right.”

Useful prompt vocabulary

Styles and mediums

Photographic	Illustration	Other
editorial photo, product shot, street photography, macro, aerial	flat vector, line art, watercolor, isometric, children’s book illustration	3D render, clay style, paper cut-out, infographic style, pixel art

Lighting

• Natural: soft window light, golden hour, overcast daylight, midday sun, blue hour.
• Studio: softbox lighting, rim light, high-key (bright), low-key (dark and moody).
• Atmospheric: backlit, dappled light through leaves, neon glow, candlelight.

Camera and composition

• Shots: close-up, medium shot, wide shot, overhead flat-lay, eye level, low angle.
• Lens feel: wide-angle, 50mm natural perspective, telephoto compression, macro detail.
• Depth: shallow depth of field (blurred background), everything in focus.
• Layout: rule of thirds, centered, negative space for text, symmetrical.

Mood and color

Calm, energetic, cozy, clinical, nostalgic, premium; muted pastels, high contrast, monochrome, earthy tones, or your brand’s colors described in words (“deep teal and warm cream”).

Natural language vs keyword prompts

Conversational generators, such as those built into ChatGPT and Gemini, handle full sentences, context and follow-up edits well: you can say “make the lighting warmer and move the mug to the left”. Some dedicated art tools respond well to shorter prompts built from comma-separated phrases. A practical rule: write a clear sentence first, then trim if the tool seems to ignore parts of it.

Prompt tips that make the biggest difference

1. Lead with the subject and purpose: “Blog header for an article about budgeting: …”.
2. Describe materials and textures: linen, brushed steel, matte ceramic, weathered wood.
3. Specify light direction and quality: it changes the whole image.
4. State the aspect ratio: choose it for where the image will be used.
5. Ask for space for text: “empty area on the left for a headline” if you will add text later.
6. Say what you want instead of what you don’t: “a quiet empty beach” rather than “no people”.
7. Use references when you can: many tools accept an image to guide style, color or composition.
8. Change one thing at a time: when refining, adjust one element per round so you learn what works.

Aspect ratios for common uses

Use	Typical ratio
Blog featured image / website header	16:9 or 3:2
Pinterest pin	2:3
Instagram feed post	4:5 or 1:1
Stories, Reels, Shorts	9:16
YouTube thumbnail	16:9
Product image	1:1

Text in images

Newer models render short text much better than older ones, but spelling mistakes and garbled letters still happen, especially with longer phrases. For anything important, such as titles on featured images, generate the image without text and add the words yourself in a design tool. If you do want text generated, put it in quotation marks and keep it short.

Keeping images consistent across a site

• Use a style block: write one paragraph describing your visual style (lighting, palette, mood, medium) and append it to every prompt.
• Reuse reference images: tools that accept image references can match a previous result’s style.
• Keep the same aspect ratio and framing for all featured images.
• Save prompts that worked alongside the images they produced.

Here is the style block we used for this site’s featured images:

Realistic editorial photograph for a blog header, 16:9, natural soft light,
teal and warm off-white color accents, shallow depth of field.
No text, letters, logos or watermarks anywhere in the image.

What we learned generating images for this site

We created the featured images on Build With AI HQ with AI image tools, using the style block above plus one sentence describing each article’s scene. A few lessons from that process:

• “No text” is not always obeyed: one design tool added a headline overlay despite the instruction, so we discarded that image.
• Screens attract logos: a laptop showing “search results” came back with a recognizable search engine logo on screen. Asking for “blurred, unreadable results” helps.
• Specific props tell the story: sticky notes grouped into clusters for keyword research, a calculator beside a laptop for a calculator guide. One concrete object per image made each one clearly match its article.
• Consistency came from the shared style block, not from using the same tool every time.

Prompt templates for bloggers and site owners

Blog featured image

Realistic editorial photo, 16:9, of [scene that represents the article's topic
with one specific prop]. [Lighting]. [Brand colors] accents. Shallow depth of field.
No text, logos or watermarks.

Pinterest pin background

Vertical 2:3 image of [subject] in [setting], bright and airy, soft natural light,
clean empty area in the top third for a headline. No text.

Flat illustration for a how-to post

Flat vector illustration of [concept], simple shapes, limited palette of
[colors], white background, friendly and modern, no text.

Product lifestyle shot

Lifestyle photo of [product description] being used by [person/context]
in [setting], natural light, [mood], product clearly visible and in focus, 4:5.

Hero image for a landing page

Wide 16:9 hero image of [scene showing the result your customer wants],
[style], [lighting], generous negative space on the left for a headline,
[brand colors] palette.

For writing better prompts in general, see how to write AI prompts, and for more blogger templates, 25 AI prompts for bloggers.

Before and after

Weak: “A beautiful image of a home office.”

Better: “A small, tidy home office by a window: oak desk, laptop, a trailing pothos plant and a ceramic mug. Soft morning light from the left, editorial interior photography, eye-level medium shot, warm neutral tones with deep green accents, calm and focused mood, 16:9.”

Weak: “Illustration about saving money.”

Better: “Flat vector illustration of a glass jar filling with coins beside a small plant growing from a coin stack, simple rounded shapes, palette of teal, cream and soft yellow, white background, optimistic mood, 1:1, no text.”

Editing and iterating

• Refine in conversation: “same image, but golden hour light” or “remove the second cup”.
• Use inpainting or edit tools: many generators let you select an area and change only that part.
• Upscale at the end: generate at the tool’s default size, pick the best, then upscale if you need a larger file.
• Finish in a design tool: crop to the exact size, add text and logos yourself, and export as WebP or compressed JPEG for the web.

Troubleshooting common problems

Problem	Try
Image looks generic or stock-like	Add specific materials, props, location and light direction
Wrong or distorted hands and faces	Regenerate, crop, choose angles where hands are less prominent, or edit that area
Garbled text or random letters	Ask for no text and add text yourself; describe screens as blurred
Ignores part of the prompt	Move the important element earlier; remove conflicting details; shorten
Unwanted objects appear	Describe the scene positively (“empty table”) or use a negative prompt field
Style changes between images	Use a fixed style block and reference images
Colors off-brand	Name colors in words and mention where they appear

For step-by-step repairs of hands, faces, text and blur, see how to fix AI image problems. To compare the tools themselves, see the best free AI image generators.

Ethics, copyright and responsible use

• Real people: do not generate realistic images of real people in misleading or harmful situations; many tools block this anyway.
• Brands and trademarks: avoid generating other companies’ logos or products in ways that suggest endorsement.
• Artists’ styles: prompting for a living artist’s name to copy their style is controversial; describe the qualities you want instead.
• Usage rights: check each tool’s terms for commercial use, and remember copyright rules for AI-generated images vary by country.
• Honesty: do not use AI images as “real” photos where that would mislead, such as fake product results, reviews or news images. Label AI images where readers would expect to know.
• Accessibility: write accurate alt text describing what the image shows.

10 ready-to-use AI image prompts

Copy these, swap the details in brackets, and keep your own style block at the end.

1. Cozy recipe header: “Overhead flat-lay of a rustic bowl of [dish] on a linen tablecloth, fresh herbs and a wooden spoon beside it, soft diffused daylight, warm earthy tones, food magazine photography, 3:2.”
2. Travel blog hero: “Wide landscape photo of [place] at golden hour, a lone traveler with a backpack seen from behind on a viewpoint, long soft shadows, vivid but natural colors, 16:9, space in the sky for a title.”
3. Finance explainer: “Clean isometric 3D illustration of a small house on a stack of coins with a rising line chart behind it, soft pastel palette of teal and cream, white background, friendly and trustworthy, 1:1, no text.”
4. Tech tutorial: “Close-up of hands on a laptop keyboard with a blurred code editor on screen, dark desk, soft blue monitor glow and warm desk lamp light, shallow depth of field, 16:9, no readable text.”
5. Fitness article: “Candid photo of a woman in her 40s stretching on a yoga mat in a bright living room, morning sunlight through sheer curtains, relaxed and encouraging mood, natural colors, 3:2.”
6. Pet care guide: “A golden retriever puppy sitting beside a food bowl on a kitchen floor, looking up at the camera, soft window light, warm and playful mood, eye-level shot, shallow depth of field, 16:9.”
7. Home improvement: “Before-and-after style split image of a small room being painted, left side bare primer walls, right side finished sage green walls with a roller tray on the floor, bright daylight, realistic photo, 16:9, no text.”
8. Business concept: “Minimal flat illustration of three people around a large lightbulb connecting puzzle pieces, simple shapes, navy and warm yellow palette, white background, optimistic, 16:9, no text.”
9. Product mockup: “Studio product photo of a [product] on a pale stone plinth, soft shadow, seamless light beige background, softbox lighting from the top left, crisp detail, premium minimal style, 1:1.”
10. Seasonal social post: “Autumn still life of a steaming mug, a knitted blanket and scattered maple leaves on a wooden windowsill, rainy window in the background, cozy moody lighting, 4:5, empty space at the top.”

Tips for popular image tools

Tool type	What works well
ChatGPT and Gemini image generation	Full-sentence descriptions, context about the purpose, and conversational edits (“keep everything, just make it evening”). Good at following detailed instructions and short text.
Midjourney	Strong aesthetics from concise descriptive prompts; its own parameters for aspect ratio and style; style and image references for consistency.
Stable Diffusion and Flux	Precise control with negative prompts, seeds, and add-ons for pose or composition; more setup, more control.
Adobe Firefly	Style, composition and structure reference controls, generative fill for editing parts of existing images, designed with commercial use in mind.
Canva AI features	Quick images inside a design workflow; add text and layout in the same place. Check that generated designs don’t add unwanted text.

Features and rules change often, so check each tool’s documentation and usage terms before relying on a specific setting.

From article to featured image in 10 minutes

1. Pick one concrete scene (1 minute): ask “what single object or moment represents this article?” For a keyword research guide: sticky notes grouped into clusters next to a laptop.
2. Write the prompt (2 minutes): scene sentence + your saved style block + aspect ratio.
3. Generate three or four options (2 minutes): pick the one with the clearest subject and fewest artifacts.
4. Refine once (2 minutes): one follow-up edit, such as warmer light or removing an object.
5. Check details (1 minute): zoom in on hands, screens, text and edges.
6. Export and optimize (2 minutes): crop to size, convert to WebP, name the file descriptively (for example ai-keyword-research.webp) and write alt text.

If you publish on WordPress, set it as the featured image and let your SEO plugin use it for social sharing. For the rest of the publishing workflow, see how to write a blog post with AI.

Image prompt cheat sheet

If you want…	Add to your prompt
A premium, calm look	“soft diffused light, muted palette, lots of negative space, minimal props”
Energy and excitement	“dynamic angle, bold saturated colors, motion blur, dramatic lighting”
Trust and professionalism	“clean composition, natural light, neutral tones, real-world setting”
Warmth and coziness	“golden hour or lamp light, warm tones, soft textures like wool and wood”
A modern tech feel	“cool tones with one accent color, reflective surfaces, subtle glow, shallow depth of field”
A friendly explainer	“flat vector illustration, rounded shapes, limited palette, white background”

Image prompt terms explained

• Depth of field: how much of the image is sharp. “Shallow” blurs the background to make the subject stand out; “deep” keeps everything in focus.
• Golden hour: the soft, warm light shortly after sunrise or before sunset. Flattering for people and landscapes.
• Blue hour: the cool, even light just before sunrise or after sunset. Good for city scenes and calm moods.
• High-key / low-key: bright, low-contrast images versus dark, high-contrast ones.
• Rim light: light from behind the subject that outlines its edges and separates it from the background.
• Flat-lay: objects arranged on a surface and photographed straight from above.
• Negative space: empty areas in the composition, useful for adding headlines later.
• Isometric: a 3D-looking illustration drawn at a fixed angle without perspective, popular for tech and explainer graphics.
• Bokeh: the soft, out-of-focus light shapes in a blurred background.
• Aspect ratio: the width-to-height shape of the image, such as 16:9 for headers or 2:3 for pins.
• Seed: a number some generators use to reproduce a similar result; reusing it with small prompt changes helps compare variations.
• Inpainting: regenerating only a selected part of an image, such as a hand or background object.
• Upscaling: increasing an image’s resolution after generation.

You do not need all of these on day one. Learning five or six, such as lighting type, shot type, depth of field and aspect ratio, will noticeably improve most prompts.

Lighting prompts: the fastest way to change the mood

Lighting changes an image more than almost any other word in your prompt. Try swapping only the lighting phrase and keeping everything else the same:

Lighting phrase	Effect	Good for
soft window light from the left	Gentle, natural, flattering	Blog headers, portraits, products
golden hour, warm low sun	Warm, nostalgic, long shadows	Lifestyle, travel, outdoor scenes
overcast daylight	Even, soft, no harsh shadows	Food, e-commerce, documentation
dramatic side lighting, deep shadows	Moody, high contrast	Editorial, bold marketing visuals
backlit, rim light around the subject	Glowing edges, separation from background	Portraits, product hero shots
neon signs at night, reflections on wet street	Urban, cinematic, colorful	Tech, gaming, nightlife themes
bright studio lighting, white seamless background	Clean, commercial	Product and catalog images
candlelight, warm dim interior	Intimate, cozy	Home, food, seasonal posts

Light fades and gradients

For soft transitions, such as light fading into darkness or a background that fades to a brand color, describe the direction and the end color: “soft light fading from the top left into deep navy in the bottom right corner” or “background gradient fading from warm cream to pale teal, subject lit softly from the front”. Gradients with plenty of empty space also leave room for headline text you add later in a design tool.

Photo editing prompts for Gemini and ChatGPT

Chat-based image tools such as Gemini and ChatGPT can edit photos you upload. Clear edit prompts say what to change and what to keep:

Change the background to a bright, minimal kitchen. Keep the person, their
clothes, pose and face exactly the same. Match the lighting to the new scene.

Make this photo look like golden hour: warmer light from the right, longer
soft shadows. Do not add or remove any objects.

Remove the people in the background and fill the space with the same street.
Keep everything else unchanged.

Turn this product photo into a lifestyle image: place the mug on a wooden
table next to an open book, soft morning light, shallow depth of field.
Do not change the mug's shape, color or logo.

Ask for one change at a time and compare the result with the original at full size, since edits can alter details you did not mention. More workflows are in AI photo editing.

Frequently asked questions

What makes a good AI image prompt?

A clear subject, action, setting, style, lighting, composition, color and format, with the most important details first and specific visual words instead of vague adjectives.

How long should an image prompt be?

Usually one to four sentences. Conversational tools handle longer prompts well; some art generators work better with shorter ones. Cut anything that does not change the image.

Can I use negative prompts?

Some tools have a separate negative prompt field for things to exclude. In plain prompts, describing what you want (“empty street”) tends to work better than “no cars”.

Why does my AI image have strange text or logos?

Models often invent text on signs and screens. Ask for no text, describe screens as blurred, and add real text yourself afterwards.

Which AI image generator is best?

It depends on your needs: conversational tools like ChatGPT and Gemini are easy to edit in plain language, while dedicated art tools offer more stylistic control. Test the same prompt in two or three tools.

Can I use AI-generated images on my blog?

Generally yes, within each tool’s terms of use. Avoid misleading uses, other brands’ logos and realistic images of real people, and add alt text.

How do I get the same style across many images?

Use one saved style description in every prompt, keep the same aspect ratio and framing, and use reference images if your tool supports them.