Relational verification establishes properties that re late two executions of a program, such as robustness, non-interference, sensitivity, and relational cost bounds. While existing techniques can verify unary safety properties of array programs, relational properties of array programs remain largely out of reach for automated tools. We present relational cell morphing, which tracks a small number of distinguished array cells across two executions simultaneously, reducing relational verification of array programs to constrained Horn clause (CHC) solving over scalar variables without requiring array theories. A dynamic cell reselection mechanism, combined with universally quantified prophecy variables, enables tight relational cost bounds parameterized by the number of positions where two arrays agree. A merged asynchronous scheduler further extends the approach to programs with different iteration structures, where type-based relational cost analysis produces only loose bounds. We evaluate on 65 encodings spanning 8 property classes, all verified in seconds to minutes by PCSat.