GPT Image
Create text-heavy layouts, product images and precise edits with GPT Image. Learn prompting, references and quality settings.
Last verified
Best for
Use OpenAI’s GPT Image models for text-heavy designs, charts, app mockups and targeted edits. Provide exact wording, data and layout instructions, then check the result for mistakes.
Sunburst is one of the slower models here. For painterly or cinematic images, try Nano Banana. For vectors, use Arrow. Which model for what shows all five families on the same prompts.
Which version
| Model | Best for | Credits per image | Reference images |
|---|---|---|---|
| GPT Image 2.6 Sunburst | OpenAI's image model for detailed layouts and precise edits | 30 | Up to 8 |
| GPT Image 2.6 Flare | OpenAI's fast image model for everyday generation and edits | 30 | Up to 8 |
Both versions have the same prices. Use Flare for new images, at about 16 seconds compared with Sunburst’s 30. Use Sunburst for edits where preserving the source matters. On GPT Image 2.5, both rendered the requested text correctly in this poster example:
GPT Image 2.5 Flare
GPT Image 2.5 Sunburst
Write the prompt
Start with the subject and purpose. For complex layouts, group the prompt into scene, subject, details and constraints. The brand board above describes its six panels in reading order.
Quote exact text and place each line. Describe the position and typography, then specify "Each line appears exactly once. No other text." In these posters, a vague request invented a date and address; quoting the copy kept those details under control.
Mentions the text
Quotes the text
Name the change and what to preserve. Attach the source image and identify the area to edit. "Make it a night scene" changed every panel below. Specifying the bottom-right photo and preserving the rest kept the other panels intact.
Repeat the details to preserve on each edit to reduce drift.
More tips
Include the data for charts and slides. Write the numbers and labels directly in the prompt. The slide above specifies every monthly value.
Describe the interface clearly. Specify the layout, spacing, hierarchy, buttons and their labels.
Give each reference a role. Identify images by number and explain what to use from each: subject, style, clothing or background.
Repeat defining character details. Attach a reference and reuse a short description of the features to preserve. For Fern’s turnaround, we specified the coat, satchel, colors and illustration style.
Specify the photographic look. Use "photorealistic" with lens, framing and lighting details. Camera terms guide appearance but do not guarantee a result.
Settings
GPT Image 2.6 has two settings besides the aspect ratio. Quality runs from Low to Max, the default is High, and it is the one that changes the price.
| Model | Credits per image | Other settings |
|---|---|---|
| GPT Image 2.6 Sunburst | 30 | Quality: Low 3, Medium 12, Extra high 59, Max 118 |
| GPT Image 2.6 Flare | 30 | Quality: Low 3, Medium 12, Extra high 59, Max 118 |
Start drafts on Low, at 3 credits per image, and increase quality when the result needs it. In this slide example, made on GPT Image 2.5, Low rendered the numbers and labels correctly; higher levels mainly changed the axis and typography.
Thinking sets how much the model plans before it draws: Low, Medium or High, and the default is High. More planning helps with dense text and complex layouts. Every level costs the same.
GPT Image offers seven aspect ratios, from 16:9 to 9:16. It has no 21:9, 5:4 or 4:5. A request for one of those, such as from an AI app, runs at the nearest ratio it has: 21:9 as 16:9, 5:4 as 4:3 and 4:5 as 3:4. The image is saved with the ratio it ran at.
Limits
- It reads up to 8 reference images.
- It doesn't make transparent backgrounds. Generate on a plain background and remove it afterwards.
- Tiny text can still come out wrong. OpenAI suggests Medium or High for small text and dense panels.
Sources
OpenAI's own guides: Image prompting and the image prompting cookbook.
Images made in Batch Banana on October 2, 2026, on GPT Image 2.5, which GPT Image 2.6 has since replaced. Select Prompt on an image to read the prompt and model that made it. Side-by-side pairs and comparison rows show the first result.