Skip to content

Images

How do I get readable text in an image?

Text inside images — headlines, labels, signs — is one of the hardest things for image models. A few habits make it much more reliable.

Pick a model that handles text

GPT Image models are the most reliable at spelling and placing text. Nano Banana Pro and Seedream also handle short text well. Purely stylistic models are more likely to garble letters.

Keep the text short and quoted

Put the exact words in quotation marks and keep them short: a headline of two to six words works far better than a paragraph. For example: a poster with the headline "Summer Sale" in bold white letters at the top. If you need more copy, add it later in a design tool.

Say where and how

Describe the position, size and style of the text: centred at the top, large bold sans-serif, white on a dark background. Clear contrast and simple fonts come out cleaner than script or decorative lettering.

Fix small mistakes with an edit

If the image is right but one letter is wrong, do not regenerate from scratch. Upload the image as a reference to an editing model and ask it to correct only the word. This keeps the rest of the image and usually costs less than repeated full generations.

When to add text yourself

For long copy, legal text, prices that must be exact, or text in many languages, generate the image without text and add the words in a design tool. It is faster and guaranteed to be correct.

FAQ

Can models write in Chinese, Japanese or Korean?
Some can, with mixed reliability. GPT Image and Seedream do best. Keep it short and check every character.
Why is the spelling right in one image and wrong in the next?
Generation is random each time. Generate a few variations at a lower cost and keep the one that is correct.
Can I make logos?
Yes, as concepts. For a final logo, redraw the chosen concept as a vector in a design tool.

Related help articles