FastSAM, a public image segmentation model trained on everyday images, is used to achieve a customizable and state-of-the-art segmentation model minimizing the training in two completely different scenarios. In one example we consider macroscopic X-ray images of the knee area. In the second example, images were acquired by microscopy of the volumetric zebrafish embryo retina with a much smaller spatial scale. In both cases, we analyze the minimum set of images required to segmentate keeping the state-of-the-art standards. The effect of filters on the pictures and the specificities of considering a 3D volume for the retina images are also analyzed.