AI tools

How to Write AI Image Prompts That Match the Picture in Your Head

By Sauravjit Singh · Updated 2026-11-15 · 6 min read

Short answer: To write a good AI image prompt, describe six things: the subject, the setting, the style or medium, the lighting, the composition and camera angle, and any text and constraints such as aspect ratio. Be concrete rather than poetic, put the most important element first, and refine with one change at a time.

Image generators do not read your mind; they turn words into visual decisions. Every detail you leave out, the model fills in with its most average guess, which is why vague prompts produce images that look like generic stock art.

An image prompt describes a finished picture rather than a task: you are briefing a photographer or illustrator. The context-and-constraints principles from our pillar guide to AI prompting still apply, but the vocabulary is visual (lenses, light, composition). Below is a formula, then one Indian scene, a Diwali product flat-lay, built version by version so you can see what each change does. This guide is for creating new images; if you want to change a photo you already have without losing the face, use our identity-safe photo editing prompts instead.

The 6-part image prompt formula

Write the parts roughly in this order. Models tend to give more weight to what comes first.

PartWhat to describeExample words
1. SubjectThe main thing, with specific details'a gift box of six soy-wax diya candles in brass holders'
2. SettingWhere it is and what surrounds it'on a dark teak table with marigold petals and a folded silk cloth'
3. Style or mediumPhoto, illustration, watercolour, 3D render'professional product photography', 'flat vector illustration'
4. LightingSource, direction, warmth, mood'warm candlelight from the lit diyas, soft fill light from the left'
5. Composition and cameraAngle, distance, framing, lens'top-down flat-lay', 'close-up', 'shot on a 50mm lens, shallow depth of field'
6. Text and constraintsExact text, aspect ratio, what to avoid'square 1:1', 'no text', 'leave empty space at the top for a headline'

You do not need all six parts every time. A quick idea sketch may need only subject and style. But when an image keeps coming out wrong, check which of the six you left out; in our experience the missing piece is usually lighting or composition, because beginners describe what is in the picture and forget how it is photographed.

One Diwali flat-lay, six versions: what each change did

Say you run a small D2C candle brand shipping from Noida and need a Diwali product image for Instagram. Here is how the prompt grows, and why each addition matters.

  1. Version 1, subject only: 'Diwali candles.' Result: random candles, often Western pillar candles, random background. The model guessed everything.
  2. Version 2, specific subject: 'A gift box of six soy-wax diya candles in small brass holders, lid open.' Now the product is recognisably yours, but the background is still a guess.
  3. Version 3, add setting: '…on a dark teak table with scattered marigold petals and a folded maroon silk cloth.' This gives Indian festive context without clichés like fireworks.
  4. Version 4, add style and lighting: '…professional product photography, lit by the warm glow of the diyas, with soft fill light from the left so the box is clearly visible.' This fixes the common problem of a dark, muddy image where only the flames show.
  5. Version 5, add composition: '…top-down flat-lay, box in the lower centre, square 1:1 format.' The image now fits an Instagram post and looks deliberate.
  6. Version 6, add constraints: '…no text, no people, empty dark space in the top third for a headline added later.' This leaves room for your offer and avoids garbled AI text.

The final prompt is long, but every phrase is doing a job. When a result is close but wrong, change one part at a time; changing four things at once makes it impossible to tell what helped.

Which words control style, lighting and camera?

A small vocabulary goes a long way. These terms are understood by almost every image model.

  • Lighting: soft window light, golden hour, overcast daylight, studio softbox, rim light, candlelight, neon, high contrast, low-key (dark and moody), high-key (bright and airy).
  • Camera distance: extreme close-up, close-up, medium shot, full body, wide shot.
  • Angle: eye level, top-down or flat-lay, low angle (looking up, makes things look grand), three-quarter view.
  • Lens feel: 35mm (natural, street), 50mm (portrait), 85mm (flattering portraits, blurred background), macro (tiny details).
  • Style: editorial photography, documentary photo, watercolour, flat vector, 3D render, pencil sketch, Madhubani-style folk art.
  • Colour: muted earthy tones, warm gold and maroon, pastel, monochrome; name two or three colours rather than 'colourful'.

AI image prompt examples you can adapt

Each example below uses the six parts. Swap the details for your own subject.

  • Blog header: 'Flat vector illustration of a small-business owner in a Delhi kirana store checking orders on a phone, three colours (teal, mustard, white), simple shapes, wide 16:9 format, empty space on the right for a title.'
  • Food menu photo: 'Close-up of a steel thali with dal makhani, jeera rice, two rotis and salad on a wooden table, overhead natural daylight, editorial food photography, square format, no text.'
  • Event poster background: 'Abstract background of soft gold bokeh lights and deep maroon gradient, festive mood, portrait 4:5, no text, no objects in the centre third.'
  • Real-estate social post: 'Bright, airy living room of a modern West Delhi builder floor with large windows, afternoon light, wide-angle interior photography, eye level, no people.'
  • Children's worksheet art: 'Cute watercolour illustration of a mango tree with a parrot, light pastel colours, white background, simple enough to colour in.'

How do you get text right in AI images?

Keep it short, quote it exactly, and say where it goes. For example: 'the words "Diwali Sale" in bold gold serif letters centred at the top'.

Short English text is now handled reasonably well by the stronger models. Long sentences, prices, phone numbers and Hindi in Devanagari still go wrong often enough that you should add them afterwards in a design tool. Our comparison of free AI image generators includes a Hindi-text test you can run to see how each tool copes.

Midjourney prompt tips vs ChatGPT and Gemini prompts

The formula is the same, but the writing style differs by tool.

MidjourneyChatGPT and Gemini
Prompt styleComma-separated descriptive phrasesFull natural sentences work well
Aspect ratioParameter such as --ar 4:5Say it in words: 'portrait 4:5 format'
Excluding things--no parameter, such as --no textSay it in the sentence: 'no text, no people'
Style strength--stylize parameter controls how artistic it getsDescribe the style in words
RefiningVariations and remix of a chosen imageConversational edits: 'make the box bigger, keep everything else'

Editing an existing photo needs a different prompt shape (what to keep, then the one change), which is why it has its own guide linked above.

How do you refine an image that is almost right?

Treat the first image as a draft and fix one thing per round. Most good images take three to five rounds.

  1. Name what is right first, so the tool keeps it: 'Keep the box, the angle and the lighting exactly as they are.'
  2. Change one element: 'Replace the marigold petals with rose petals.'
  3. If the tool keeps drifting, restate the full final prompt with the fix built in, rather than adding a tenth follow-up.
  4. Save every version you like. Models rarely reproduce an exact image twice.
  5. When it is close enough, finish small fixes (a stray petal, a crooked line) in an editor instead of regenerating.

Common image prompt mistakes

  • Poetic adjectives instead of visual facts. 'A magical, stunning, breathtaking Diwali' gives the model nothing concrete. 'Warm candlelight, marigold petals, brass' does.
  • Stacking conflicting styles: 'photorealistic watercolour anime' confuses most models.
  • Forgetting the aspect ratio, then cropping out the important part for Instagram or a banner.
  • Asking for real people, celebrities or other brands' logos. Avoid this entirely for business use.
  • Writing a new prompt from scratch after every miss instead of changing one part.
  • Not stating the setting, so Indian scenes drift into generic or foreign-looking backgrounds.

Prompting for images, video and voice is covered hands-on in our 8-week AI course, with no coding involved.

Frequently asked questions

How long should an AI image prompt be?

Usually one to four sentences, or about 30 to 80 words. Long enough to cover the six parts; beyond that, extra words often get ignored or create conflicts.

What is a negative prompt?

A negative prompt lists what you do not want in the image, such as text, extra people or blur. Midjourney uses the --no parameter; in ChatGPT and Gemini you simply write 'no text, no people' in the prompt.

Do the same prompts work in every AI image generator?

The six-part content transfers well, but the format differs. Midjourney prefers short phrases and parameters, while ChatGPT and Gemini respond well to full sentences and conversational edits.

How do I keep the same product or person across several images?

Upload a reference image and ask the tool to keep the subject identical while changing only the setting or angle. Tools built on image editing, such as Gemini's image model, are generally better at this consistency.

Can I use AI image prompts for commercial work?

Most tools allow commercial use of generated images, but terms vary and change. Check the current terms, and avoid real people, logos and copyrighted characters.

Want to learn this with a trainer?

Leave your details and our counsellor will call you within one working day with batch dates, fees and the right course for you.

  • Free counselling call, no obligation to join
  • Weekday and weekend batches, online or offline
  • Small batches of 15 to 20 students

Or chat on WhatsApp

Enter a 10-digit mobile number.

We use your details only to contact you about Spettro School courses.

Chat on WhatsApp