Article

How to write an AI image prompt: method and examples

Write an effective AI image prompt: subject, setting, composition, light and constraints. Copy the template, adapt three examples and test it on Deepnia.

Deepnia Editorial Team · · 7 min read

Steaming coffee cup on the left of a wooden table, lit by a window, with open space on the right.
AI-generated editorial illustration showing a clear subject, specific lighting and intentional empty space. This is not an output tested in Deepnia.

An effective AI image prompt describes a visible subject, explains how it should be framed and identifies the details that matter for the final use. Start with a clear scene, then add lighting, visual treatment and a few essential constraints. After generating, revise the instruction that relates to the main problem rather than rewriting everything.

Test your prompt in Deepnia Image with Nano Banana 2.

The goal is not to find a secret phrase that makes every image look professional. It is to give the model a coherent brief that you can evaluate. For the wider process, including settings and export, see the beginner’s guide to creating AI images.

Start with the decision the image must support

Before describing colors or camera angles, finish this sentence: “The reader should understand that…” For a product announcement, the answer might be that a new reusable bottle is available. For a tutorial cover, it might be that a rough idea can become a finished illustration.

That purpose determines what belongs in the picture. A bottle announcement needs a recognizable bottle. A tutorial about planning may work better with a desk and a small number of visual notes. An attractive background is not useful if it competes with the main message.

Keep this purpose in your project notes. You do not always need to include it word for word in the generation prompt, but it should guide every visible choice. When deciding whether to add another object, ask whether it clarifies the message or merely decorates the frame.

Use five building blocks

1. Subject and action

Name what the image is about, and say what it is doing when that matters. “A person working” is broad. “An adult arranging paper color samples on a desk” gives the scene a recognizable activity. If there should be one subject, state that instead of leaving the number open.

2. Setting

Place the subject somewhere specific enough to support the task. “A tidy home office with a plain wall” supplies context without demanding dozens of background details. If the setting does not matter, use a simple background and spend your attention on the subject.

3. Composition

Describe the viewpoint and arrangement. Useful directions include close-up, overhead view, centered object or subject on the left. Reserve space for a later headline when needed. Do not request a tight close-up and a complete room view in the same image unless you explain a deliberate multi-panel layout.

4. Light and visual treatment

Choose a treatment that fits the purpose: photographic, watercolor, paper collage or a restrained editorial illustration. Then describe the light or color behavior in ordinary language. “Soft window light with gentle shadows” is more actionable than a pile of adjectives such as stunning, premium and incredible.

5. Essential constraints

Identify the few details that would make the output unusable if ignored. You might need no lettering, a plain background or the complete object inside the frame. Constraints should protect the brief, not become an exhaustive list of every defect you have ever seen.

A reusable prompt worksheet

Complete these fields before turning them into a short paragraph:

  • Purpose: Where will the image appear?
  • Subject: What must be recognizable immediately?
  • Setting: What context genuinely helps?
  • Composition: Where should the subject and any empty space go?
  • Treatment: What medium, lighting and palette fit the task?
  • Constraints: Which two or three requirements are non-negotiable?

You can leave a field out when it does not matter. This is a thinking aid, not mandatory syntax. A long prompt is not automatically more precise.

Turn the worksheet into this adaptable starting point:

Show [subject and action] in [setting]. Frame it from [viewpoint], with [placement and space for later layout]. Use [light] and [visual treatment]. Keep [essential details] and avoid [one or two unwanted elements].

Replace the brackets with concrete descriptions, then remove any irrelevant sentence.

Put your brief into practice

Open Deepnia Image when your worksheet is ready. Adapt one of the examples below and test it with Nano Banana 2.

Example 1: a blog cover with room for a headline

Suppose your article explains how to organize a creative project. A weak request is “Make a beautiful productivity image.” It leaves the subject, message and layout unresolved.

Try this example prompt instead:

Create an editorial illustration of an adult organizing five colored paper cards on a wooden desk. Use a slightly overhead view. Place the hands and cards in the lower-left part of the frame and leave the upper-right area visually quiet for a headline added later. Use a limited navy, cream and muted-yellow palette, with soft shadows. No readable words or brand logos.

If the number of cards is essential to explaining a process, count them in the result. If it is merely decorative, do not keep regenerating to fix an irrelevant detail.

Example 2: a fictional product concept

For an invented product, define shape and material:

Create a photographic-style concept image of one unbranded reusable water bottle. The bottle is matte forest green, cylindrical, with a simple screw cap. Show the complete bottle standing upright on a neutral stone surface, at eye level, with a soft shadow. Keep the background warm gray and uncluttered. Do not include lettering, packaging or accessories.

For a real product, start from its photo in Product Shooting; the AI product photo guide shows how.

Example 3: changing a visual treatment

In Deepnia Image, upload the source image and explain what should change and what should remain:

Turn the supplied fictional café image into a watercolor-style illustration. Preserve the cup’s position, the table layout and the empty space on the right. Replace the photographic surface treatment with visible watercolor washes and soft paper texture. Keep the warm palette and do not add new objects or lettering.

This separates style from content. If the model changes the cup or fills the empty area, decide which error matters most before requesting another revision.

Diagnose a weak result before adding more words

If the picture is too busy, simplify the object list or explicitly prioritize one subject. If the viewpoint is wrong, clarify composition rather than adding quality adjectives. If the colors clash, reduce the palette. If the picture looks polished but communicates the wrong idea, return to the original purpose.

Some fixes sit outside the prompt: set the aspect ratio in the Deepnia Image options, and use a sharper reference if details drift.

Use a small revision record: original request, observed problem, one change, outcome. For example: “Too little headline space → move subject further left.”

Know when to stop revising

Define a usable result before generating. For the cover example, that could mean a clear activity, a coherent hand position and enough room for a readable headline. Once those conditions are met, move to layout and export instead of searching indefinitely for a slightly different aesthetic.

Save the chosen file together with the prompt and settings. If the work will be repeated later, keep a short explanation of the decisions that matter. “Space for a two-line headline” is a more useful production note than “make it look like the good one.”

Frequently asked questions

Should prompts always be in English?

No. Write clearly in the language you know best, such as English or French, and rephrase any ambiguous term.

Do camera and lens terms improve every prompt?

No. They are useful when you understand the appearance you want to describe. Ordinary wording such as “close-up with a softly blurred background” may express your intention more clearly than unfamiliar specifications.

Should I use a negative prompt?

You don’t need one: state two or three exclusions in the prompt itself, for example “no lettering, no extra objects.”

Your next prompt

Choose one scene, complete the worksheet and check that every sentence serves the same purpose. Then open Deepnia Image, paste your brief and compare two models, such as Nano Banana 2 and GPT Image 2; for 4K brand visuals, try Nano Banana Pro. If the result needs attention, use the guide to fixing an AI image without starting over rather than replacing the whole brief immediately.

To pick a model for your project, see how to choose an AI image generator. To organize repeat work, read Building an AI image creation workflow.

Turn this method into your first creation

Pick your tool, paste your prompt and launch your first creation.

Create my accountSee the tools