OCR.now exposes only language packs actually installed on the server. A single selected model can create a draft, but mixed scripts may need separate passes and deliberate recombination.
Only installed language packs are offered. The workflow helps decide whether one model preserves the page or whether regions must be recognized and recombined separately.
Set the limits of this task with How to Choose an OCR Language Model, then use A Reviewable Workflow for Mixed-Language OCR to review uncertainty. Related inputs may be better handled by Image to Text or Handwriting to Text.
When separate language passes are safer
Use separate script-specific recognition when one model consistently damages the other script. Choose a specialist or human transcription workflow for unsupported languages, historical scripts, or legally consequential multilingual documents.
Test one combined pass against script-specific regions
Identify every script and reading direction before recognition.
Run separate crops when one language model damages the other script.
Keep names and untranslated terms in their source form during review.
Inspect transitions between languages and directions
| Review point | Likely failure | Validation action |
|---|---|---|
| Script coverage | The chosen model lacks one writing system | Inventory every script and confirm an installed language pack |
| Direction | Right-to-left and left-to-right runs reorder | Check copied text, numbers, punctuation, and region sequence |
| Names | Dictionary expectations replace proper nouns | Protect names and compare them character by character |
Mixed-script page scenario
For a bilingual notice with English headings and Chinese body text, test a combined page pass and focused script-specific crops. Compare which method preserves names, numbering, and line order. Recombine by visible region rather than translating or rewriting text to make the output sound fluent.