In case 6, letters from the English alphabet were used as targets. Unlike all other test images used in this work, the letters were hand-drawn, making their existence in the design space completely unknown. Due to the top–bottom symmetry in sculpted flow shapes, we are only able to design for symmetric letters. Since the flow physics are entirely contained within the transition matrices, future optimization problems could incorporate new geometry that allows for top–bottom asymmetry, such as half-pillars, or “steps” in the microchannel. Here, we drew the capitalized four letters in “IOWA,” choosing the axis of symmetry in the microchannel to align with similar axes in the letters. The images were resized for three different microchannel aspect ratios: *h*/*w* = [0.25, 0.5, 1.0], where *h* is the height of the microchannel. In searching across microchannel aspect ratios, we expect the designed shapes to have considerably different styles of deformation. Amini et al. [1] established that changing the microchannel aspect ratio will introduce new modes of fluid deformation, with some modes being unique to particular aspect ratios. The GA searches used nine channels for *h*/*w* = 1.0 and seven channels for *h*/*w* = [0.25, 0.5]. The results for *h*/*w* = 0.25 are in Fig. 11, and the results for *h*/*w* = 0.5 and 1.0 are in Appendix B (Fig. 13). Matching fluid flow shapes are found for each letter, despite the unique flow deformations available across different channel aspect ratios. This is a highly useful result, as manual design would require memorization of the various modes for each aspect ratio in addition to combinatorial difficulty. All results show that while the framework is effective in designing for bulk fluid displacement and overall shape, subtle features are still difficult to optimize for. As an example, the hollow portion of the letter “A” is never truly found, though the outline of the letter does not seem difficult. This indicates room for improvement on the fitness function itself, which remains a simple correlation function. Promising avenues include using elliptic Fourier representations of shape, wavelet representation, or using methods from image classification like scale-invariant feature transforms.