(1) We randomly selected three conformations of the target protein and identified three potential epitope regions following prior work. (2) Based on the preprocessed targets, we applied four versions of our sequence–structure binder co-design models—JointDiff-binder and JointDiff-x-binder—to generate candidate sequences with sizes of 100, 150, 200, and 250. For each combination of target, model, and size, we designed five sequences, resulting in a total of 720 sequences. (3) We evaluated the novelty and diversity of the designs. All sequences were sufficiently novel, with pairwise sequence identities below 0.3. (4) For each sequence, we used Boltz-2 to predict structures and the corresponding iPAE, and then computed the ipSAE. We generated three predictions per sequence and used the average ipSAE as the final score. (5) We ranked the sequences based on these scores. (6) Finally, we selected the top 100 sequences as our final submission.