Check the words.
Keep the evidence.
Compare text on your device screens with your translation table, review differences and keep a traceable record.
1. Add your files
Select your screenshots and translation table together, or use the separate “Upload translation table” button. Choose the language shown on the device.
- Images: PNG, JPG and JPEG.
- Translation tables: XLSX and CSV. PDF, Word and Numbers files are not supported; export your table to XLSX or CSV first.
- Up to 100 files per selection, with a 20 MB limit for each file. Failed files can be retried separately.
Check the detected worksheet and language columns. Advanced mapping lets you correct the selection. Numbers export-summary sheets are skipped when a suitable translation sheet is detected.
2. Compare against a reference
Start the analysis after checking your table and target language. A reference selected for comparison is not automatically treated as customer-approved. Only mark a baseline approved when you have that approval.
You can pause a run and continue unfinished images later. Poor-quality photos and ambiguous text remain available for human review.
3. Review and export
Review each image beside its detected text and reference translation. Use the controls below the image to change pages or zoom. Confirm a difference, correct the recognition, or dismiss a false positive.
Export review reports in CSV, XLSX or HTML. The device correction list includes confirmed display differences. To verify a repair, upload new evidence; the original image cannot prove that a repair was made.
Frequently asked questions
Does AI recognition keep my images on this computer?
No. In this Cloudflare pilot, AI recognition runs remotely: Cloudflare sends the image and target language to OpenRouter, which routes the request to Google Gemini. The translation table is not included in the recognition request. See the data-processing notes for more details.
Can a matching result guarantee a correct translation?
No. StringProof compares visible text with your reference. Recognition can miss or misread text, and a reference may itself contain errors. Review evidence before approving a result.
Why are some entries marked as needing review?
Reflections, clipping, blur, uncertain recognition or multiple possible table entries prevent a reliable automatic decision. Correct the text, select the appropriate reference, or upload a clearer image.
Does it detect text missing from a screen?
Only when you supply an expected-text list that identifies what should appear on that screen. Without it, unused table entries are not evidence of missing device text.
What happens when I delete a task?
The application removes its stored task data, original files and exports. Offline backups have a separate retention policy and are not erased by deleting an online task. Ask your workspace operator about that policy.
Create an account with Google. Each Google account receives 20 successful AI image analyses over its lifetime; failed or canceled analyses do not count. Imports, manual review, translation versions and exports remain available. Service-wide reliability protections may temporarily pause AI analysis even when your account still has allowance remaining. No payment or paid checkout is available. Interface in six languages.