What we'd really like is a PII redaction model for video...