feat: Implement Character Memory V1 - VLM analysis and prompt injection

api/routers/editor.py

@@ -473,8 +473,35 @@ async def regenerate_frame_image(
         else:
             logger.warning(f"[REGEN-IMG] No style anchor found for {storyboard_id}")
         
-        # Apply style prefix to prompt
-        final_prompt = f"{style_prefix}, {prompt}" if style_prefix else prompt
+        # Get character descriptions for prompt injection
+        character_prefix = ""
+        from api.routers.quality import _character_stores
+        if storyboard_id in _character_stores:
+            char_descriptions = []
+            for char_data in _character_stores[storyboard_id].values():
+                appearance = char_data.get("appearance_description", "")
+                clothing = char_data.get("clothing_description", "")
+                name = char_data.get("name", "character")
+                
+                if appearance or clothing:
+                    parts = [f"{name}:"]
+                    if appearance:
+                        parts.append(appearance)
+                    if clothing:
+                        parts.append(f"wearing {clothing}")
+                    char_descriptions.append(" ".join(parts))
+            
+            if char_descriptions:
+                character_prefix = "Characters: " + "; ".join(char_descriptions) + ". "
+                logger.info(f"[REGEN-IMG] Injecting character descriptions: {character_prefix[:80]}...")
+        
+        # Apply style prefix and character descriptions to prompt
+        final_prompt = ""
+        if style_prefix:
+            final_prompt += f"{style_prefix}, "
+        if character_prefix:
+            final_prompt += character_prefix
+        final_prompt += prompt
         logger.info(f"[REGEN-IMG] Final prompt: {final_prompt[:120]}...")
         
         # Use MediaService to generate image via RunningHub workflow