Improve side portrait quality by using original face reference

Enhanced Stage 2 (side portrait generation) to preserve facial details:
- Added original face image as second reference input
- Updated prompt to instruct AI: "Use the details on the second image to fill in missing parts from the first image"
- Preserves side details (ears, hair, jawline) that may be obscured in frontal view

Changes only affect side portrait generation (Stage 2).
All other stages remain unchanged.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

character_forge_image/services/character_forge_service.py

@@ -195,12 +195,17 @@ class CharacterForgeService:
             if progress_callback:
                 progress_callback(2, current_stage)
 
-            # Original prompt - works for ALL backends
-            current_prompt = f"Create a side profile view of this {gender_term}{costume_instruction} focusing on the face filling the entire available space. The {gender_term} should be shown from the side (90 degree angle) with professional studio lighting against a neutral grey background. Maintain exact facial features from the reference images."
+            # For side portrait, add original face to preserve side details
+            # Prompt instructs AI to use details from original (second image) to fill in missing parts from front portrait (first image)
+            current_prompt = f"Create a side profile view of this {gender_term}{costume_instruction} focusing on the face filling the entire available space. The {gender_term} should be shown from the side (90 degree angle) with professional studio lighting against a neutral grey background. Maintain exact facial features from the reference images. Use the details on the second image to fill in missing parts from the first image."
 
-            input_images = [front_portrait, reference_full_body]
+            # Always include front_portrait as first image, original face as second
+            input_images = [front_portrait, reference_face]
+
+            # Add additional references based on mode
             if initial_image_type == "Face + Body (Separate)":
-                input_images.append(reference_face)
+                # Already have both front_portrait and reference_face
+                pass
             elif costume_image:
                 input_images.append(costume_image)