Understand image-tool credit use

Image operations are priced per image. Generation and improvement estimates depend on the selected model and output settings as well as the prompt and source images.

The unit is one image

Selecting several images means a separate charge for each successful operation. There is no batch discount, and each image is processed as its own request.

What determines the estimate

Generation and improvement offer four models: Nano Banana 2 (Google Gemini), Nano Banana Pro (Google Gemini), ChatGPT Sunburst (OpenAI) and ChatGPT Flare (OpenAI).

The estimate takes account of:

• Model and resolution. Nano Banana 2 has separate 1K and 2K prices. Nano Banana Pro has the same base output price at 1K and 2K. Only sizes within 2560 pixels per edge are offered; Google 2K is unavailable at 9:16, 16:9 and 21:9.
• OpenAI quality. Low, Medium, High, XHigh and Max have different costs. High is the default.
• OpenAI dimensions. The aspect ratio and resolution together determine the pixel dimensions used for the estimate.
• Complete prompt. This includes the chosen style and supplied business context.
• Input images. Improvement includes the original and any reference images.
• Number of outputs. Each image is a separate operation.

The estimate is approximate and credits are rounded up to a whole number. Changing the settings can change the price. Choosing whether to replace the original or create a new attachment does not change the image generation charge.

Background removal and metadata generation have their own pricing and settings.

Estimating before a batch

Use the submit option that calculates credits before confirming. Check the model, resolution and quality, then confirm that the selected image count matches your intended batch.

Start with one image at a lower resolution. Review it at its intended display size before increasing quality or processing the rest. Request a new estimate whenever you change the settings.

What is free

• Searching and browsing the media library.
• Selecting and deselecting images.
• Choosing styles, models, ratios and options.
• Requesting an estimate.
• Cancelling the confirmation.
• Cropping, resizing or deleting an image afterwards.
• Editing generated metadata by hand.

Planning a large project

A catalogue background-removal pass. Count the products, choose the background settings, estimate one batch and multiply. Then process in reviewable batches rather than all at once — you want the option to stop after twenty if the quality is not right.

A media library metadata pass. Batches are capped at 48 images. Do not process the whole library: prioritise images on pages that matter and skip decorative images, which should have empty alt text anyway.

A photography standardisation pass. Test on one image, then one small batch, then the rest. Testing costs one or two charges and prevents a large run with the wrong prompt.

Failed operations

Refused requests are not charged. Individual failures within a batch are not charged, while the images that succeeded in that batch are.

After any partially failed run, read the processing summary and check Review your credit history rather than assuming an all-or-nothing outcome.

Controlling spend across sites

On a shared subscription, image batches are the operation most likely to consume the pool unexpectedly. Set a per-domain spending limit in My Account → POM AI Management before anyone runs a catalogue job.

See Plans, credits and domain capacity and Which actions consume credits.