The paper surveys recent advances in generative pipelines that produce and optimize personalized advertising creatives across image and poster formats. The study synthesizes evidence on constraint ingestion, layout-aware rendering, retrieval-assisted staging, human-feedback inspection, CTR-oriented reward conditioning, and serving-time selection/ranking. Particular attention is paid to how knowledge-augmented vision-language adapters improve brand/price text handling; how poster systems encode hierarchy for legibility; how retrieval narrows the feasible space before diffusion; how inspector-driven feedback reduces unusable variants while correlating with live engagement; and how joint or parallel ranking architectures preserve creative diversity without latency penalties. The goal is to develop an operational blueprint for customer acquisition that reduces idea-to-launch cycles while maintaining brand safety and persuasive clarity. Methods employed include a comparative synthesis of ten recent studies, a structured content analysis, and a concept mapping of failure modes, evaluation metrics, and optimization objectives. The findings consolidate an end-to-end stack that aligns offline screening with online lift and reallocates human effort from repetitive triage to governance and experiment design.