GPT Image 2.5: Prompt Guide

Overview

GPT Image 2.5 renders a frame from a prompt, or edits up to 16 images you attach. This walkthrough makes one image from scratch, then changes a single detail in it.

Before you start

• Decide the shape: 3:4 for posters and covers, 16:9 for a hero, 1:1 for a packshot. Auto lets the model choose.
• Write down any words that need to appear in the image, exactly as they should read.

1. Write the frame

Describe the subject, the light, the material and the grade. Put the words you want in quotes and give each one a role and a position.

2. Set the shape and quality

Choose a fixed shape, then render it at 1K, 2K or 4K. The size menu stays off while the shape is Auto. Explore at Low quality, which comes back fastest.

3. Change one thing

Attach the image you made and describe the single change — swap the flower, melt the butter, paint the rim. Say that the framing, light, palette and grain stay identical.

4. Finish the file

Re-render the keeper at High or above. For a cutout, set Background to Transparent and save it as PNG or WebP.

Next steps

Find prompts to start from in GPT Image 2.5: Prompt Guide.