This study provides a preform design approach for uniform strain distribution in forging products based on a convolutional neural network (CNN). The appropriate preform design prevents underfill problems by improving the material flow inside forging dies and achieving a uniform strain distribution in forging products. The forging deformation process and mechanical properties are improved with a uniform strain distribution. The forging and strain distribution results are analyzed through rigid–plastic finite element forging simulations with different initial geometries. The simulation data are fed into the CNN model as an input array, from which the geometric characteristics are extracted by convolution operations with filters (weight array). The extracted features are linked to the considered initial shapes, which are input into the CNN model as an output array. The presented model derives the preform shape for a target forging with uniform strain distributions using the training weights. According to the training database, the proposed design method can be applied to different forging geometries without any iterations. By creating a number of low-level CNN (LC) models based on the training data, the efficiency of the preform design can be improved. The best preform among the derived preform candidates is chosen by comparing the forging results. Compared with previous studies using the same design criteria, the proposed model predicted the preform with a strain distribution improved by 16.3–38.4%.