"""enrich_metadata — Nova AI metadata enrichment.
Two tiers:
- premium (default): Full 8-section Golden Codex via Nova/Gemini 2.5 Pro
- standard: SEO-optimized title/description/keywords/alt_text via Nova-Lite
"""
import base64
import httpx
import structlog
from ..mcp_server.config import config
logger = structlog.get_logger()
def enrich_metadata(
image_b64: str,
context: str = "",
artist_name: str = "",
title: str = "",
copyright_holder: str = "",
creation_year: str = "",
soul_whisper: dict | None = None,
) -> dict:
"""Analyze an image and generate Golden Codex metadata.
Args:
image_b64: Base64-encoded image (PNG or JPEG).
context: Creator's brief — technical/artistic context for AI analysis.
artist_name: Artist/creator name (embedded in final metadata).
title: Artwork title (leave blank for AI to suggest one).
copyright_holder: Copyright owner name.
creation_year: 4-digit creation year.
soul_whisper: Optional personal message dict:
{"message": "...", "sender": "..."}
Returns:
dict with full Golden Codex metadata JSON.
"""
logger.info("enrich_metadata", context=context[:80] if context else "none")
# Detect MIME type from base64 header
mime_type = "image/png"
if image_b64[:20].startswith("/9j/"):
mime_type = "image/jpeg"
# Build Nova-compatible payload
payload = {
"image_data": image_b64,
"mime_type": mime_type,
"user_id": "mcp_agent",
"parameters": {
"analysis_depth": "full",
},
}
# Build context string matching Nova's user_metadata_context format
# This gets appended to the AI prompt by Nova
context_parts = []
if context:
context_parts.append(f"Creator's Brief: {context}")
if title:
context_parts.append(f'Title: "{title}"')
if artist_name:
context_parts.append(f'Artist: "{artist_name}"')
if copyright_holder:
context_parts.append(f'Copyright Holder: "{copyright_holder}"')
if creation_year:
context_parts.append(f"Creation Year: {creation_year}")
if soul_whisper and soul_whisper.get("message"):
from datetime import datetime, timezone
sw_date = datetime.now(timezone.utc).isoformat()
sw_sender = soul_whisper.get("sender", artist_name or "Anonymous")
context_parts.append('SoulWhisper ENABLED: Set "enabled": true in soulWhisper object')
context_parts.append(f'SoulWhisper Message: "{soul_whisper["message"][:1000]}"')
context_parts.append(f'SoulWhisper Sender: "{sw_sender}"')
context_parts.append(f'SoulWhisper Date: "{sw_date}"')
# Pass as context field — Nova will use it in the prompt
if context_parts:
payload["context"] = "\n\nUSER-PROVIDED METADATA (USE THESE VALUES EXACTLY):\n" + "\n".join(context_parts)
payload["context"] += '\n\nIMPORTANT: If SoulWhisper fields are provided above, you MUST set "enabled": true in the soulWhisper object of your JSON output.'
response = httpx.post(
f"{config.nova_url}/enrich",
json=payload,
timeout=180.0,
)
response.raise_for_status()
data = response.json()
result = {
"metadata": data.get("golden_codex", data),
"model": data.get("model_used", "gemini-2.5-pro"),
"schema_version": "1.1.0",
"processing_time_ms": data.get("processing_time_ms"),
}
# Ensure soul_whisper is in metadata if provided but Nova missed it
if soul_whisper and soul_whisper.get("message"):
gc = result["metadata"]
if isinstance(gc, dict) and "soulWhisper" not in gc:
gc["soulWhisper"] = {
"enabled": True,
"message": soul_whisper["message"][:1000],
"sender": soul_whisper.get("sender", artist_name or "Anonymous"),
"date": sw_date,
}
return result
# Nova-Lite URL (standard tier)
_NOVA_LITE_URL = "https://nova-lite-agent-172867820131.us-west1.run.app"
def enrich_metadata_standard(
image_b64: str,
content_type: str = "artwork",
) -> dict:
"""Generate SEO-optimized standard metadata via Nova-Lite.
Returns: title, description, keywords, alt_text — suitable for XMP/IPTC.
Costs 1 GCX (half the price of premium).
Args:
image_b64: Base64-encoded image (PNG or JPEG).
content_type: 'artwork' or 'photo' — affects analysis style.
"""
logger.info("enrich_metadata_standard", content_type=content_type)
mime_type = "image/png"
if image_b64[:20].startswith("/9j/"):
mime_type = "image/jpeg"
payload = {
"image_data": image_b64,
"mime_type": mime_type,
"user_id": "mcp_agent",
"parameters": {
"analysis_depth": "standard",
"content_type": content_type,
},
}
response = httpx.post(
f"{_NOVA_LITE_URL}/enrich",
json=payload,
timeout=120.0,
)
response.raise_for_status()
data = response.json()
# Nova-Lite returns {title, description, keywords, alt_text} in golden_codex
metadata = data.get("golden_codex", data)
return {
"tier": "standard",
"metadata": metadata,
"model": data.get("model_used", "gemini-2.5-flash"),
"processing_time_ms": data.get("processing_time_ms"),
"fields": ["title", "description", "keywords", "alt_text"],
"note": "Standard tier: SEO-optimized XMP/IPTC fields. Use tier='premium' for full Golden Codex museum-grade analysis.",
}
