The optimization goal of e‑commerce ad creatives has shifted from pure aesthetics to business metrics such as click‑through rate (CTR). Existing approaches build CTR‑based rewards and apply reinforcement learning to refine prompts, generating appealing ad images directly from product photos. This generation‑centric pipeline penalizes all attributes of a low‑performing image, overlooking attributes that may actually boost clicks.\ \ We introduce ECC (Evolving Creatives with Clicks), reformulating creative optimization as fine‑grained editing rather than generation from scratch. ECC selectively edits attributes that hurt CTR while preserving those that contribute positively. To pinpoint influential factors among massive images, we first identify 28 attributes strongly correlated with CTR using ANOVA and Random Forest regression. Based on attribute contrasts between images with disparate CTRs, we instruction‑tune a prompt model to produce editing suggestions, then refine these suggestions with real click preferences because reference‑derived hints are not uniformly beneficial.\ \ To mitigate overfitting in direct preference optimization (DPO), ECC proposes Consistent Preference Optimization (CPO), which enforces consistency across semantically equivalent shuffled suggestions, improving the generalization of the preference model.\ \ Extensive experiments demonstrate that ECC markedly improves ad image CTR, outperforming state‑of‑the‑art baselines by a large margin. The full data pipeline, attribute extraction and analysis code, the resulting dataset, and source code will be released upon acceptance.\ \ Review