Fix inconsistent photorealistic compliance severity

Add explicit guidance that photorealistic product/marketing images are
acceptable and should not be flagged as violations. Only flag photorealistic
content involving fake people or misleading contexts as ERROR severity.

Fixes QA bug where photorealistic issues were randomly flagged as critical
or warning.

Author: hunterjam

content-gen/src/backend/settings.py

@@ -312,6 +312,14 @@ def get_compliance_prompt(self) -> str:
 - Do not generate images depicting violence, weapons, or harmful activities
 - Avoid culturally insensitive or appropriative imagery
 
+**IMPORTANT - Photorealistic Product Images Are ACCEPTABLE:**
+Photorealistic style for PRODUCT photography (e.g., paint cans, products, room scenes, textures) 
+is our standard marketing style and should NOT be flagged as a violation. Only flag photorealistic 
+content when it involves:
+- Fake/deepfake identifiable real people (SEVERITY: ERROR)
+- Misleading contexts designed to deceive consumers (SEVERITY: ERROR)
+Do NOT flag photorealistic product shots, room scenes, or marketing imagery as violations.
+
 ### Compliance Validation
 The Compliance Agent MUST flag any content that violates these RAI principles as SEVERITY: ERROR.
 RAI violations are non-negotiable and content must be regenerated.