Practical guides
How to write AI image prompts
Describe what must appear, where it belongs and what the image is for. Generate a first version, then make focused edits while preserving what works.
On this page
- 1. Build the prompt from five useful parts
- 2. Choose the format in Image Studio
- 3. Use a reusable image prompt
- 4. Worked example: a fictional ceramic mug campaign
- 5. Use references and edit the result you want to keep
- 6. Handle text and transparent backgrounds deliberately
- 7. Review before downloading and publishing
- Common questions
- Sources
A useful AI image prompt describes the subject, setting, composition, visual style and important constraints. Choose the image shape in the app’s controls, generate a first version, then review it against the brief. For an edit, attach the source image and state both what should change and what must stay.
You do not need a long string of impressive adjectives. A clear instruction such as “place one blue mug on the right, with empty space on the left for a headline” gives you a result you can evaluate. Decide the purpose before describing the scene.
1. Build the prompt from five useful parts
- Purpose: A website banner, product concept, presentation illustration or invitation background.
- Subject and setting: What appears, how many subjects there are, and where the scene takes place.
- Composition: The viewpoint, placement, crop and space needed for other content.
- Visual treatment: Photography, watercolor, flat illustration or another coherent medium, plus lighting and palette.
- Constraints: Required details, exact text if any, and unwanted elements that would make the result unusable.
OpenAI’s image prompting guide recommends describing composition and specifying what an edit should preserve. Treat prompt details as instructions to check, rather than guarantees of exact geometry, typography or subject identity.
Avoid conflicting requests. “A tightly cropped face with the entire body visible” asks for incompatible framing. Replace it with the view you actually need. If realism matters, describe the visible effect, such as soft window light and natural skin texture, instead of relying on camera jargon alone.
2. Choose the format in Image Studio
Open Image Studio. In Settings, choose Image and the aspect ratio that suits the destination: Square 1:1, Landscape 3:2 or Portrait 2:3. These are the current single-image choices. If you need a different final crop, leave room around the subject and check it in your destination editor.
You can start with No style and describe the visual treatment in the prompt, or choose a matching style preset. Avoid selecting a preset that contradicts your written brief. Auto is available as a model choice for generation; when reference images are attached, image-edit model selection is automatic.
The shape control and the written composition serve different purposes. Choosing Landscape does not tell the model to leave space for a title. Put that placement requirement in the prompt as well.
3. Use a reusable image prompt
Create a [visual medium] for [intended use]. Show [subject, number and important details] in [setting]. Composition: [viewpoint, subject placement and empty space]. Lighting and color: [specific treatment]. Keep [required features]. Include only this text, if any: [exact wording and placement]. Avoid [unwanted elements]. The image should communicate [one clear idea].
Delete fields you do not need. For an illustration without text, say “no lettering”. For a scene with several objects, describe their relationships: the notebook lies in front of the mug, the lamp stands behind it. A list of objects alone leaves their arrangement open.
4. Worked example: a fictional ceramic mug campaign
Fictional example: Cedar Kiln is an invented pottery brand preparing a concept image for a new blue mug. The brief needs a warm landscape image with room for a headline. There is no real product photo attached, so the result is a visual concept, not an accurate product photograph.
Vague prompt: “Beautiful premium mug advertisement, amazing lighting, high quality.”
Useful prompt:
Create a photographic concept for a handmade pottery campaign. Show one matte blue ceramic mug with a rounded handle on a pale oak table. Place the mug in the right third of a landscape composition, viewed slightly from above. Leave the left half mostly empty for a headline that will be added later. Use soft morning light from the left, gentle shadows and a warm cream background. Keep the whole mug and handle visible. No lettering, logos, steam, flowers or extra cups.
This brief gives each review question a clear answer: one mug or several, full handle or cropped, empty left side or cluttered. It also separates the image from the headline, which can be added as editable text later.
If the mug is attractive but too central, keep the successful visual treatment and edit only the placement. Starting again with unrelated adjectives makes it harder to know which change helped.
5. Use references and edit the result you want to keep
For an existing product, person or design, attach a suitable reference photo with the paperclip. Explain its role: “Use this mug’s shape and blue glaze; change only the background.” A reference supplies visible information that words may not capture, but you still need to inspect the result for altered details.
To revise a generated image, open its preview and choose Edit image, or use Refine on the result. This carries the selected image into the next edit. Confirm that the intended reference is attached before submitting your correction.
Move the mug farther right so the left half is clear. Preserve its blue color, shape, handle, lighting and table surface. Keep the whole mug visible. Do not add text or extra objects.
Make one substantial change at a time and compare with the previous version. Google’s image generation documentation describes image inputs and conversational editing; the exact controls and limits you can use still depend on the app.
6. Handle text and transparent backgrounds deliberately
If you need text inside the image, supply the exact words in quotation marks and describe their position. Keep the amount manageable. Zoom in to check every character, date and number. A convincing poster at thumbnail size can contain misspellings or extra lettering.
For information that changes often, such as prices or event dates, consider generating the background and adding the final text in a layout editor. That gives you editable wording without regenerating the illustration each time.
To cut out an existing subject in Image Studio, attach the image and use Remove background in Settings. The app offers this as a transparent PNG workflow. Inspect fine edges, hair, handles and shadows in the downloaded file. A white background or a painted checkerboard is not proof of transparency.
7. Review before downloading and publishing
Pair product imagery with a description grounded in the actual product specifications. Neither the image nor the copy should imply a feature the item does not have.
- Composition: Check the subject count, crop and space required for headings or buttons.
- Details: Inspect hands, handles, object connections, repeated patterns and any brand-specific features.
- Text: Read all visible lettering at full size, including small marks you did not request.
- Reference fidelity: Compare shapes, colors and recognizable features with the original image.
- Destination: Check the downloaded dimensions, transparency and crop in the page, slide or design where it will appear.
Use Download image from the preview when you have a version worth keeping. Keep the original reference and the prompt with your project so later edits have a clear starting point. If the image represents a real product, verify it against that product before presenting it as an accurate depiction.