How to Write AI Image Generation Prompts (5-Part Formula + Examples)
Last updated: October 2026
To write a good AI image prompt, describe five things: your subject, the scene, the visual style, the lighting and camera angle, and the final image format.
A few specific details can make your instructions much clearer. This guide shows you how to write AI image generation prompts, improve them step by step, and fix common problems in ChatGPT and Gemini.
Key Takeaways
- Use the 5-part formula: Subject + Scene + Style + Lighting & Camera + Format.
- Describe visible details: Replace vague words like "beautiful" with specific colors, materials, poses, and lighting.
- Improve one detail at a time: Keep what works and tell the AI exactly what to change.
Why Your AI Image Prompts Don't Work
When your prompt leaves important details open, the AI fills them in. A request like "a beautiful room" gives the model freedom to choose the furniture, lighting, colors, and camera angle.
Three problems are especially common:
| Problem | What Happens | How to Fix It |
|---|---|---|
| Too vague | The image feels generic | Describe the subject and setting |
| Missing visual direction | The framing or style is wrong | Specify lighting, composition, and style |
| Conflicting instructions | Important details get lost | Prioritize one clear visual direction |
Some problems also depend on the image model. Complex poses, exact text placement, and keeping several characters consistent may require additional attempts or reference images.
The 5-Part AI Image Prompt Formula
A useful AI image prompt reads like a short creative brief. Start with these five elements and add only the details your image needs.
| Part | What to Describe | Example |
|---|---|---|
| 1. Subject | Who or what appears? | A fluffy white Samoyed puppy |
| 2. Action & Setting | What's happening, and where? | Playing in a puddle on a Tokyo street |
| 3. Style | What should it look like? | Realistic lifestyle photography |
| 4. Lighting & Camera | How is it lit and framed? | Soft evening light, low-angle close-up |
| 5. Format & Constraints | What shape and details matter? | Vertical 3:4, full body visible |
You can use this structure for portraits, product photos, illustrations, and more. The most useful details depend on what you're creating.
Copy-and-Paste AI Image Prompt Formula
Prompt Template
Here's a completed example.
Copy Prompt
The prompt specifies the puppy's appearance, activity, environment, lighting, and framing. Each detail helps communicate the image you want.
How to Write AI Image Prompts in 4 Steps
Let's build one image prompt from a simple idea: a woman sitting in a café.
Step 1: Start With the Main Subject
Write one short sentence describing what you want to see.
This gives the AI a subject and a location. Her outfit, expression, pose, lighting, and surroundings are still open to interpretation.
Step 2: Add the Setting and Visual Style
Give the scene more personality by describing the outfit, surroundings, and photographic style.
Revised Prompt
Now the prompt describes the outfit, setting, and atmosphere more clearly.
Step 3: Add Lighting, Composition, and Format
Tell the AI how to frame the subject and where the light should come from.
Detailed Prompt (Copy & Paste)
The detailed prompt makes the intended outfit, pose, lighting direction, and composition easier to understand.
You can use the same approach for product photography. For example, "a perfume bottle" can become "a pink glass perfume bottle with a gold cap on a cream marble pedestal, soft studio lighting, realistic reflections, and a centered 1:1 composition."
Step 4: Generate, Review, and Refine
Check your generated image against the original idea. Find the biggest detail that needs improvement.
For example, if the portrait is too tightly cropped, use this follow-up:
Follow-Up Prompt
Changing one detail at a time helps you evaluate what worked without losing the parts you already like.
Want to practice with a finished portrait idea? Start with an existing prompt and customize the outfit, pose, or background.
How to Edit an Existing Image With AI
When editing an uploaded photo, your prompt should explain what to change and what must stay the same.
Use this structure:
Reference Image + Change + Preserve + Desired Result
Example 1: Change a Hairstyle Without Changing the Face
Upload a clear photo of yourself, then use:
Photo Editing Prompt (Copy & Paste)
Example 2: Redesign a Living Room
Upload your room photo, then use:
Photo Editing Prompt (Copy & Paste)
How to Use Multiple Reference Images
If your image generator accepts multiple uploaded images, assign each one a specific role.
For example:
Reference Image Prompt
This tells the AI which details belong to each image. It can be especially useful for outfit previews, hairstyle changes, and creative photo transformations.
ChatGPT vs Gemini: How to Write Better Image Prompts
ChatGPT and Gemini both support natural-language instructions for generating and editing images. The same five-part formula works as a starting point in either tool.
Their official prompting guides emphasize similar fundamentals, with slightly different examples and workflows.
| Task | ChatGPT | Gemini |
|---|---|---|
| Generate an image | Describe the subject, intended use, composition, and constraints | Describe the subject, action, location, composition, and style |
| Edit an uploaded photo | Separate the changes from the details to preserve | Use direct editing instructions and identify reference-image roles |
| Add text | Quote exact wording and specify typography and placement | Quote exact wording and describe typography and placement |
| Refine a result | Reuse the image and restate important constraints | Make targeted changes through conversational follow-ups |
Writing AI Image Prompts for ChatGPT
OpenAI's image prompting guidance recommends describing the result you want, including its intended use, subject, composition, style, and important constraints.
For example:
ChatGPT Prompt
For editing, be explicit about what should stay unchanged. If you upload several images, explain what each reference contributes to the final result.
Writing AI Image Prompts for Gemini
Google's Gemini prompting guidance highlights subject, composition, action, location, and style. Additional details such as lighting, camera angle, and reference-image instructions help communicate more specific visual goals.
For example:
Gemini Prompt
For image editing, try direct instructions such as "change only the dress color to burgundy" while identifying the details that should remain consistent.
Both tools can interpret descriptive prompts and follow-up edits. Results depend on the image model, available controls, and the complexity of your request.
How to Fix Common AI Image Prompt Problems
If your generated image is close to what you imagined, start with the biggest mismatch before rewriting everything.
| Problem | Try This Prompt Fix |
|---|---|
| The subject is too small | "Move the camera closer and use a medium shot." |
| The full body is cropped | "Zoom out and show the entire subject, including both feet." |
| The lighting feels flat | "Use warm side lighting with natural shadows." |
| The background is distracting | "Simplify the background to a clean cream wall." |
| The face changes during editing | "Preserve the recognizable facial features from the uploaded reference." |
| The hands look unnatural | "Use a relaxed pose with naturally positioned hands." |
| Unwanted objects appear | "Keep only the main subject and the specified objects." |
| The colors don't match | "Use a consistent palette of cream, beige, and warm brown." |
| Image text is incorrect | "Use the exact text 'SLOW MORNINGS' and no additional words." |
These instructions may improve the next result, but complex details can still require several attempts.
How to Get Text Right in AI-Generated Images
If you're creating a poster, thumbnail, or social media graphic, put the exact words you want in quotation marks.
Also describe the text's position, size, and style.
Copy Prompt
Keep the wording short and proofread the result. For designs that require exact brand typography, a final pass in a design editor may still be useful.
Should You Use Negative Prompts?
Start by describing the result you want. Add direct exclusions when they clarify an important requirement.
For example, "a clean product image with no text or additional objects" is a reasonable instruction for ChatGPT or Gemini.
Some other image generators include dedicated negative-prompt controls. Use the settings supported by your chosen tool.
Where to Find Ready-Made AI Image Prompts
Writing your own prompts gives you creative control. Starting from a ready-made prompt can also help you discover visual styles and learn which details produce the look you want.
Browse a collection of AI image prompts, choose a visual idea, and customize the subject, lighting, outfit, or background.
You can also explore themed prompts for seasonal photos, fashion portraits, product photography, or interior design. Seeing how different prompts describe their subjects and settings is a useful way to improve your own writing.
Frequently Asked Questions
How long should an AI image generation prompt be?
One to three clear sentences can be enough for a simple scene. Complex compositions and photo edits may need longer instructions. Include the details that matter without adding unnecessary descriptions.
How do I write a prompt for realistic AI images?
Describe a believable subject, natural lighting, camera framing, and realistic materials. For portraits, mention natural skin texture and a physically plausible pose. For product images, specify accurate materials and reflections.
Can I use the same prompt in ChatGPT and Gemini?
Yes, you can start with the same subject, setting, style, and lighting instructions in both tools. Review the results and adjust the wording if one model interprets a detail differently.
How do I keep the same face in AI-generated images?
Upload a clear reference photo and ask the image generator to preserve recognizable facial features, skin tone, and natural proportions. Reuse the reference when possible and check each result for identity changes.
Can I write AI image prompts on my phone?
You can use supported image-generation features in the ChatGPT and Gemini mobile apps. Type or paste your prompt into the chat, and upload a reference photo if you want to edit an existing image. For longer prompts, drafting them in a notes app first can make editing easier.
What's the easiest way to improve a bad AI image?
Identify one problem, such as incorrect lighting, framing, or background. Ask the AI to change that detail while preserving everything that already looks right.
Final Thoughts
Learning how to write AI image generation prompts starts with describing your ideas clearly and refining them deliberately.
Use the five-part formula, generate an image, and check what needs improvement. Over time, you'll discover which details matter most for your favorite visual styles.
Ready to create something new? Choose a ready-made image prompt, customize the details, and make the idea your own.