Train your own model using your own field of view images or go back to DeepStack.
Remember AI is only as good as the images the model has to go with. So maybe between your field of view and what images were used for each model can cause the differences.