Generate an image

The image generator creates a new picture from a description and uploads it to your media library. It is the safest POM AI tool to learn on, because it only ever adds.

Open it at Media → AI generate images.

The form

The screen shows the remaining credit balance, then four cards.

Describe your image. The prompt, up to 3,100 characters, with a counter below showing what remains. An example prompt is prefilled so you can see the expected level of detail. A template dropdown appears above it if you have saved templates for the image generator context, along with a Dictate prompt control.

Choose a style. One preset, applied on top of your prompt:

• Realistic portrait
• Product photography
• Realistic landscape
• Web illustration
• 3D illustration
• Marvel Comic style
• None (only based on your prompt)

Choose a model. Select Nano Banana 2 (Google Gemini), Nano Banana Pro (Google Gemini), ChatGPT Sunburst (OpenAI) or ChatGPT Flare (OpenAI). Nano Banana 2 is the default. All four offer a Resolution choice. OpenAI models also show Quality (OpenAI), defaulting to High, and the exact pixel dimensions for the selected aspect ratio. See Image formats and dimensions before choosing larger output.

Select an aspect ratio. Ten options, from 21:9 through 1:1 to 9:16, defaulting to 16:9.

Running it

Use the button that calculates credits before confirming, at least until you know what your usual settings cost.

After you accept, a status line reports that the image is being generated and counts the elapsed seconds. Generation is not instant. Do not resubmit — a second submission is a second paid image.

The result

The generated image appears under Your generated images at the bottom of the screen, and has already been uploaded to the media library. It is available anywhere an attachment can be used.

Nothing else on your site is modified. The generator does not attach the image to a post, does not set a featured image and does not replace anything.

Choosing settings sensibly

Style before prompt. The presets do a lot of work. Picking Product photography achieves more than a paragraph describing studio lighting. Use None only when you want the prompt to control everything.

Start with one image. Check the credit estimate at 1K, review the composition, and compare another model or higher quality only when the result needs it.

Aspect ratio to match the slot. Choose the ratio the image will actually occupy: 16:9 or 21:9 for a hero band, 1:1 for a card or thumbnail, 4:5 or 9:16 for mobile-first placements. Generating at the right ratio avoids a crop that loses the subject.

Iterating

Image prompting is iterative. Expect two or three attempts to reach something usable, and budget for that.

To keep the cost of iteration down:

• iterate at a lower resolution and check the estimate for your selected model;
• change one thing per attempt, so you learn what caused the difference;
• keep the prompt that worked — save it as a template under Settings → POM AI → Prompt templates.

See Write a useful image prompt.

Before using a generated image publicly

• Look at it properly, full size. Generated images have artefacts — hands, text, reflections, repeated patterns — that are invisible in a thumbnail.
• Check for text. Generated lettering is usually wrong. If the image needs words, add them yourself.
• Check nobody recognisable appears. Do not present a generated likeness as a real person, and do not use it where a real photograph is implied.
• Check for logos and trademarks that drifted into the picture.
• Write real alternative text. See Write alternative text that helps.
• Consider whether a generated image is honest here. A product photograph must show the product a customer receives.

Review generated and edited images covers this in full.

Cost

One charge per generation, based on the model, resolution, OpenAI quality and complete prompt. See What image operations cost.

When to use a different tool

• To improve a photograph you already have, use AI improve image.
• To cut out a subject, use AI remove image background.
• To describe images you already have, use AI metadata for images.