A good text-to-image prompt describes six things: the subject, what it’s doing, the setting, the style or medium, the lighting and the composition. “A cat” gives you a generic cat. “A ginger cat asleep on a stack of old books in a sunlit library, watercolor, soft morning light, close-up from a low angle” gives you a picture you had in mind. The order matters less than being specific.
This guide covers the structure, 20 example prompts you can copy and how prompting differs between ChatGPT, Gemini, Midjourney and open models like Stable Diffusion and FLUX.
The six parts of an image prompt
| Part | What to describe | Example |
|---|---|---|
| Subject | Who or what, with 2 or 3 specific details | “an elderly fisherman with a white beard and a yellow raincoat” |
| Action | What the subject is doing | “mending a net” |
| Setting | Where and when | “on a wooden pier in a Norwegian harbor at dawn” |
| Style or medium | Photo, oil painting, 3D render, pencil sketch, flat illustration | “35mm film photograph” |
| Lighting and mood | Light source, time of day, atmosphere | “low golden light, light mist, calm” |
| Composition | Framing and camera angle, aspect ratio | “medium shot, eye level, wide 16:9” |
Put together: “35mm film photograph of an elderly fisherman with a white beard and a yellow raincoat, mending a net on a wooden pier in a Norwegian harbor at dawn, low golden light, light mist, medium shot at eye level, 16:9.”
Tips that make the biggest difference
- Be concrete. “Red leather armchair” beats “nice chair”. Models fill vague words with the most average version.
- Name the medium. “Photograph”, “gouache illustration”, “isometric 3D render” changes the whole image.
- Describe light. Lighting does more for mood than adjectives like “beautiful” or “stunning”.
- Use the model’s language for text in images. If you need words in the image, put them in quotes: a sign reading “OPEN LATE”.
- Change one thing at a time. When refining, edit a single detail so you can see its effect.
- Avoid real people and brands unless you have rights. Most tools restrict them, and publishing such images can create legal problems.
20 example prompts by style
Photography
- “Close-up photograph of a barista pouring latte art, shallow depth of field, warm café light, steam rising, 50mm lens.”
- “Aerial drone photo of a winding mountain road through autumn forest, high contrast, early morning fog.”
- “Studio product photo of a ceramic coffee mug on a pale green background, soft shadow, minimal, centered.”
- “Street photograph of a rainy Tokyo crosswalk at night, neon reflections on wet asphalt, umbrellas, cinematic.”
Illustration and art
- “Children’s book illustration of a shy hedgehog carrying a lantern through a snowy forest, soft watercolor, warm palette.”
- “Flat vector illustration of a home office with plants and a laptop, pastel colors, clean lines, no text.”
- “Oil painting of a lighthouse in a storm, thick brush strokes, dramatic sky, in the style of 19th-century maritime painting.”
- “Pencil sketch of an old city tram, cross-hatching, sketchbook page, slight paper texture.”
- “Japanese woodblock print of a fox under a cherry tree, limited palette, bold outlines.”
3D and design
- “Isometric 3D render of a tiny bakery, cutaway view showing the kitchen and shop, soft lighting, toy-like.”
- “App icon of a paper plane on a blue gradient, rounded square, simple, high contrast.”
- “Clay-style 3D character of a friendly robot gardener, studio lighting, white background.”
- “Minimal poster for a jazz festival with the text "BLUE NOTES 2026", Bauhaus style, red, cream and black.”
Concept and fantasy
- “Concept art of a floating market on a giant lily pad in a misty swamp, lanterns, wide shot, painterly.”
- “A library inside a hollow ancient tree, spiral staircase, glowing mushrooms, cozy, detailed.”
- “Retro-futuristic city in 1960s space-age style, monorail, sunny day, optimistic, illustration.”
Practical uses
- “Simple diagram-style illustration showing the water cycle, labeled arrows, flat colors, for a school worksheet.”
- “Blog header image of a desk with a notebook and a cup of tea, top-down view, soft natural light, empty space on the left for text.”
- “Mood board of a Scandinavian living room: light wood, linen, beige and sage tones, four images in a grid.”
- “Realistic photo of a modern kitchen renovation with dark green cabinets and brass handles, wide angle.”
How prompting differs by tool
ChatGPT and Gemini
ChatGPT (with ChatGPT Images 2.5, released in September 2026) and Gemini (with Google’s Nano Banana models) understand full sentences and conversation. You can describe the image naturally, then refine with follow-ups: “make it evening”, “remove the second chair”, “keep everything but change the coat to blue”. Both can also edit an image you upload. They are the easiest place to start.
A note on older guides: OpenAI retired the separate DALL·E GPT inside ChatGPT on 30 August 2026, so “DALL·E prompts” now simply means prompts for ChatGPT’s built-in image generation (OpenAI).
Midjourney
Midjourney (V8 family, with V8.2 released in July 2026) responds strongly to style words and has parameters added at the end of the prompt, such as an aspect ratio (--ar 16:9) or excluded elements (--no text) (Midjourney updates). Shorter, more visual prompts often work better than long sentences. Check Midjourney’s documentation for the current parameter list, since it changes between versions.
Stable Diffusion, FLUX and other open models
Open models running locally (such as FLUX.2 in ComfyUI) give you the most control: negative prompts that list what to avoid, seeds for repeatable results, and settings for steps and guidance. They reward precise, comma-separated descriptions. See self-hosted AI image generators for setup.
Use AI to write your prompts
Chatbots are good prompt writers. Try: “I want an image of [idea]. Ask me five questions about style, mood and composition, then write three detailed prompts I can use.” This is often faster than an “image prompt generator” app, and you can tailor it to the tool you use.
Fixing common problems
| Problem | Fix |
|---|---|
| Image looks generic | Add specific details for subject, setting and light |
| Wrong number of objects | State the number and position: “two cups, one on each side” |
| Garbled text | Put exact text in quotes, keep it short, or add the text later in an editor |
| Odd hands or faces | Regenerate, or crop and edit that area |
| Style keeps drifting | Repeat the style phrase in every prompt, or use the tool’s style reference feature |
FAQ
What is a text-to-image prompt?
It’s the written description you give an AI image generator. The model turns the text into an image, so the more specific the description of subject, setting, style, light and composition, the closer the result to what you imagined.
How long should an image prompt be?
One to three sentences is usually enough. Very long prompts can confuse some models; focus on the details that matter most.
Can I use AI-generated images commercially?
It depends on the tool’s terms and your country’s law. In the US, purely AI-generated images aren’t protected by copyright, so others may be able to reuse them. Read the terms of the tool you use.
What’s the best free AI image prompt generator?
Any free chatbot (ChatGPT, Claude, Gemini) can write image prompts. Ask it to question you about style and mood first.
Conclusion
Describe subject, action, setting, style, light and composition, then refine one detail at a time. Start in ChatGPT or Gemini for natural-language prompts, and move to Midjourney or open models when you want more control. To choose a tool, see the best AI image generators.