Skip to main content
Korean characters in a generated image sometimes come out smudged or replaced with different characters. Here is why it happens and what you can do on each plan.

​
1. Cause: Korean characters are harder for AI to draw

Korean characters have more strokes than letters or numbers, so drawing them accurately inside an image is harder. The longer the phrase, the more likely it breaks. Results vary a lot by model, and the high performance image models available on paid plans render Korean text far more reliably. No model is perfect every time, but the difference in success rate is clear.
If you upload an image that already contains text and ask for an edit, the original text is preserved. This issue mainly appears when AI has to draw new characters into the image.

​
2. Paid plans: specify a model that handles Korean well

Example prompt: “(Enter desired content) Use Nanobanana Pro”
High performance image models available on paid plans, such as Nanobanana Pro, render Korean text far more reliably. No model is perfect every time, so some errors can remain when the phrase is long or contains a lot of characters.

​
3. Free plan: raise your success rate on the default model

The free plan can only use the Carat default image model. If you specify another model, you will see a notice that it is available on paid plans. These approaches help on the default model:
  • Ask for shorter text in the image. Longer sentences break more often.
  • Upload an image that already has the text and ask for an edit. The original text is preserved.
  • Generate the background or image without text, then add the wording yourself.
If you regularly need results with Korean text, subscribing to a paid plan gives you access to the high performance image models that handle Korean better.

Models available on each plan

See which image and video models the free and paid plans can use.
We keep adopting newer models to improve Korean text rendering.