Feedback via Attention-Based Reference Image Conditioning