Skip to content
Contact us
  • There are no suggestions because the search field is empty.

How to Translate Images with Prism, Pairaphrase’s Image Translator AI Agent

Prism is Pairaphrase’s Image Translator AI agent. It is available in the AI Sandbox and is designed to translate text inside images.

Prism uses a two-step workflow:
  1. Create a draft translation of the image
  2. Review and edit the detected text, then generate the final image

This helps users refine translated text before producing the finished image.

How to open Prism

  1. Open the AI Sandbox in Pairaphrase.
  2. Under Translation, select Prism — Image Translator.

This opens the Prism image translation screen.

What Prism supports

Prism currently supports these image file types:

  • PNG
  • JPG
  • WEBP
  • GIF

Set up the image translation

Before translating, choose the Source Language and Target Language, then select the AI Engine you want Prism to use.

If needed, use the arrow button between the language fields to swap the source and target languages.

You can also enter optional guidance in the Instructions field. For example, you might tell Prism to keep brand names in English, use a formal tone, or leave product codes unchanged.

Once the settings are ready, upload the image you want to translate.

Note: At the time of writing, we recommend Gemini for image translation because it currently tends to produce the best results in Prism. Performance may vary by image and language pair.

Upload the image

Select Upload image — PNG, JPG, WEBP or GIF and choose the image you want to translate.

After upload, the filename appears in the upload area. You can click it again if you want to replace the file before translating.

Step 1: Create the draft translation

After selecting the language pair, AI engine, and any optional instructions:

  1. Upload the image
  2. Click Translate Image

Prism then creates a draft translated image and opens the editing interface.

Review and edit the draft

After the draft is created, Prism displays an editing screen with:

  • The original image
  • The draft translated image
  • A list of detected text blocks
  • Editable fields for the source text and translated text

This is the review stage of the workflow.

What you can edit

You can edit:

  • The detected source text
  • The proposed translated text

As you make changes, the draft preview updates so you can review the translation before generating the final image.

This is useful when you want to:

  • Correct OCR or detected text
  • Refine wording
  • Adjust terminology
  • Improve readability
  • Standardize labels or headings

Step 2: Generate the final image

Once the text looks correct:

  1. Review all translated text blocks
  2. Make any needed edits
  3. Click Generate final image

Prism then creates the final translated image.

The final image is generated using the edited text and is designed to match the original image as closely as possible, including layout, styling, and background.

Download the final image

After the final image has been generated, click Download final image to save it.

Start over with a new image

If you want to translate a different image, click New image and repeat the process.

Tips for better results

For best results:

  • Choose the correct source and target languages before starting
  • Use the Instructions field when certain text should remain unchanged
  • Review all detected text blocks carefully
  • Correct any text-detection errors before generating the final image
  • Check the translated image for layout, readability, and terminology before using it

Frequently asked questions

What is Prism?

Prism is Pairaphrase’s Image Translator AI agent in the AI Sandbox. It translates text inside images and includes a two-step review workflow.

What file types can I upload?

Prism currently supports PNG, JPG, WEBP, and GIF files.

Can I choose the AI engine?

Yes. Use the AI Engine menu to choose from the AI models available in Prism.

What does the Instructions field do?

The Instructions field lets you guide the translation. For example, you can tell Prism to keep brand names in English or avoid translating certain codes.

What happens when I click Translate Image?

Prism creates a draft translation of the image and opens the editing interface so you can review and edit the text before creating the final image.

Can I edit the translated text before generating the final image?

Yes. Prism includes an editing mode where you can review and update the detected source text and translated text.

When can I download the image?

You can download the result after the final image has been generated by clicking Download final image.

Should I review the image before downloading it?

Yes. Always review the detected text, translated text, and draft preview before generating the final image.


Need Help?

If you’d like tips on writing better prompts or using the Sandbox for specific projects, contact us here.