Local OCR before deciding whether cloud translation is allowed
Route `local-ocr-ai-translate` (translate_image.hybrid) fits teams that want local text extraction before deciding whether online translation is acceptable—e.g., vendor screenshots with internal codes. Mask sensitive tokens first, then translate only de-identified sentences. Maintain a glossary so product names do not drift across languages.
Hybrid image-translation policy tips
- Redact internal ID columns before upload; crop just the sentence you need to communicate externally.
- Confirm numerals and proper nouns after OCR, then translate; never forward unchecked screenshots.
- Log original snippets, redaction notes, translations, and timestamps inside the ticket for audit trails.
Image translation FAQ (hybrid)
How should teams define content that must never hit a cloud translator in `local-ocr-ai-translate`?
Security and legal jointly publish a deny list (IDs, secrets, contract values); default to masking or placeholder text first.
Which fields are denied by default for cloud MT?
Government IDs, secrets, internal codenames, undisclosed financials—maintain a living deny list.
Do placeholders confuse translators?
Yes—document placeholder semantics in a glossary and restore controlled truth in final docs.
How do we audit external translation calls?
Log actor, timestamp, source hash, and model revision inside the ticket system.
Is hybrid slower?
It trades speed for compliance; whitelist low-risk content for a fast lane.